Computer Networking A Top-Down Approach 6th Edition

Computer Networking A Top-Down Approach 6th Edition
COMPUTER
NETWORKING
SIXTH EDITION
A Top-Down Approach
James F. Kurose
University of Massachusetts, Amherst
Keith W. Ross
Polytechnic Institute of NYU
Boston Columbus Indianapolis New York San Francisco Upper Saddle River
Amsterdam Cape Town Dubai London Madrid Milan Munich Paris Montréal Toronto
Delhi
Mexico City
São Paulo
Sydney
Hong Kong
Seoul
Singapore
Taipei
Tokyo
Vice President and Editorial Director, ECS:
Marcia Horton
Editor in Chief: Michael Hirsch
Editorial Assistant: Emma Snider
Vice President Marketing: Patrice Jones
Marketing Manager: Yez Alayan
Marketing Coordinator: Kathryn Ferranti
Vice President and Director of Production:
Vince O’Brien
Managing Editor: Jeff Holcomb
Senior Production Project Manager:
Marilyn Lloyd
Manufacturing Manager: Nick Sklitsis
Operations Specialist: Lisa McDowell
Art Director, Cover: Anthony Gemmellaro
Art Coordinator: Janet Theurer/
Theurer Briggs Design
Art Studio: Patrice Rossi Calkin/
Rossi Illustration and Design
Cover Designer: Liz Harasymcuk
Text Designer: Joyce Cosentino Wells
Cover Image: ©Fancy/Alamy
Media Editor: Dan Sandin
Full-Service Vendor: PreMediaGlobal
Senior Project Manager: Andrea Stefanowicz
Printer/Binder: Edwards Brothers
Cover Printer: Lehigh-Phoenix Color
This book was composed in Quark. Basal font is Times. Display font is Berkeley.
Copyright © 2013, 2010, 2008, 2005, 2003 by Pearson Education, Inc., publishing as
Addison-Wesley. All rights reserved. Manufactured in the United States of America. This
publication is protected by Copyright, and permission should be obtained from the publisher prior to any prohibited reproduction, storage in a retrieval system, or transmission
in any form or by any means, electronic, mechanical, photocopying, recording, or likewise. To obtain permission(s) to use material from this work, please submit a written
request to Pearson Education, Inc., Permissions Department, One Lake Street, Upper
Saddle River, New Jersey 07458, or you may fax your request to 201-236-3290.
Many of the designations by manufacturers and sellers to distinguish their products are
claimed as trademarks. Where those designations appear in this book, and the publisher was
aware of a trademark claim, the designations have been printed in initial caps or all caps.
Library of Congress Cataloging-in-Publication Data
Kurose, James F.
Computer networking : a top-down approach / James F. Kurose, Keith W. Ross.—6th ed.
p. cm.
Includes bibliographical references and index.
ISBN-13: 978-0-13-285620-1
ISBN-10: 0-13-285620-4
1. Internet. 2. Computer networks. I. Ross, Keith W., 1956- II. Title.
TK5105.875.I57K88 2012
004.6—dc23
2011048215
10 9 8 7 6 5 4 3 2 1
ISBN-10:
0-13-285620-4
ISBN-13: 978-0-13-285620-1
About the Authors
Jim Kurose
Jim Kurose is a Distinguished University Professor of Computer Science at the
University of Massachusetts, Amherst.
Dr. Kurose has received a number of recognitions for his educational
activities including Outstanding Teacher Awards from the National
Technological University (eight times), the University of Massachusetts, and
the Northeast Association of Graduate Schools. He received the IEEE Taylor
Booth Education Medal and was recognized for his leadership of
Massachusetts’ Commonwealth Information Technology Initiative. He has
been the recipient of a GE Fellowship, an IBM Faculty Development Award,
and a Lilly Teaching Fellowship.
Dr. Kurose is a former Editor-in-Chief of IEEE Transactions on
Communications and of IEEE/ACM Transactions on Networking. He has
been active in the program committees for IEEE Infocom, ACM SIGCOMM,
ACM Internet Measurement Conference, and ACM SIGMETRICS for a
number of years and has served as Technical Program Co-Chair for those
conferences. He is a Fellow of the IEEE and the ACM. His research interests
include network protocols and architecture, network measurement, sensor
networks, multimedia communication, and modeling and performance
evaluation. He holds a PhD in Computer Science from Columbia University.
Keith Ross
Keith Ross is the Leonard J. Shustek Chair Professor and Head of the Computer
Science Department at Polytechnic Institute of NYU. Before joining NYU-Poly in
2003, he was a professor at the University of Pennsylvania (13 years) and a
professor at Eurecom Institute (5 years). He received a B.S.E.E from Tufts
University, a M.S.E.E. from Columbia University, and a Ph.D. in Computer and
Control Engineering from The University of Michigan. Keith Ross is also the
founder and original CEO of Wimba, which develops online multimedia
applications for e-learning and was acquired by Blackboard in 2010.
Professor Ross’s research interests are in security and privacy, social networks,
peer-to-peer networking, Internet measurement, video streaming, content distribution
networks, and stochastic modeling. He is an IEEE Fellow, recipient of the Infocom
2009 Best Paper Award, and recipient of 2011 and 2008 Best Paper Awards
for Multimedia Communications (awarded by IEEE Communications Society). He
has served on numerous journal editorial boards and conference program committees, including IEEE/ACM Transactions on Networking, ACM SIGCOMM, ACM
CoNext, and ACM Internet Measurement Conference. He also has served as an
advisor to the Federal Trade Commission on P2P file sharing.
iii
This page intentionally left blank
To Julie and our three precious
ones—Chris, Charlie, and Nina
JFK
A big THANKS to my professors, colleagues,
and students all over the world.
KWR
This page intentionally left blank
Preface
Welcome to the sixth edition of Computer Networking: A Top-Down Approach. Since
the publication of the first edition 12 years ago, our book has been adopted for use at
many hundreds of colleges and universities, translated into 14 languages, and used
by over one hundred thousand students and practitioners worldwide. We’ve heard
from many of these readers and have been overwhelmed by the positive response.
What’s New in the Sixth Edition?
We think one important reason for this success has been that our book continues to offer
a fresh and timely approach to computer networking instruction. We’ve made changes
in this sixth edition, but we’ve also kept unchanged what we believe (and the instructors and students who have used our book have confirmed) to be the most important
aspects of this book: its top-down approach, its focus on the Internet and a modern
treatment of computer networking, its attention to both principles and practice, and its
accessible style and approach toward learning about computer networking. Nevertheless, the sixth edition has been revised and updated substantially:
• The Companion Web site has been significantly expanded and enriched to
include VideoNotes and interactive exercises, as discussed later in this Preface.
• In Chapter 1, the treatment of access networks has been modernized, and the
description of the Internet ISP ecosystem has been substantially revised, accounting for the recent emergence of content provider networks, such as Google’s. The
presentation of packet switching and circuit switching has also been reorganized,
providing a more topical rather than historical orientation.
• In Chapter 2, Python has replaced Java for the presentation of socket programming. While still explicitly exposing the key ideas behind the socket API, Python
code is easier to understand for the novice programmer. Moreover, unlike Java,
Python provides access to raw sockets, enabling students to build a larger variety
of network applications. Java-based socket programming labs have been
replaced with corresponding Python labs, and a new Python-based ICMP Ping
lab has been added. As always, when material is retired from the book, such as
Java-based socket programming material, it remains available on the book’s
Companion Web site (see following text).
• In Chapter 3, the presentation of one of the reliable data transfer protocols has
been simplified and a new sidebar on TCP splitting, commonly used to optimize
the performance of cloud services, has been added.
• In Chapter 4, the section on router architectures has been significantly updated,
reflecting recent developments and practices in the field. Several new integrative
sidebars involving DNS, BGP, and OSPF are included.
viii
Preface
• Chapter 5 has been reorganized and streamlined, accounting for the ubiquity of
switched Ethernet in local area networks and the consequent increased use of
Ethernet in point-to-point scenarios. Also, a new section on data center networking has been added.
• Chapter 6 has been updated to reflect recent advances in wireless networks, particularly cellular data networks and 4G services and architecture.
• Chapter 7, which focuses on multimedia networking, has gone through a major
revision. The chapter now includes an in-depth discussion of streaming video,
including adaptive streaming, and an entirely new and modernized discussion of
CDNs. A newly added section describes the Netflix, YouTube, and Kankan video
streaming systems. The material that has been removed to make way for these
new topics is still available on the Companion Web site.
• Chapter 8 now contains an expanded discussion on endpoint authentication.
• Significant new material involving end-of-chapter problems has been added. As
with all previous editions, homework problems have been revised, added, and
removed.
Audience
This textbook is for a first course on computer networking. It can be used in both
computer science and electrical engineering departments. In terms of programming
languages, the book assumes only that the student has experience with C, C++, Java,
or Python (and even then only in a few places). Although this book is more precise
and analytical than many other introductory computer networking texts, it rarely
uses any mathematical concepts that are not taught in high school. We have made a
deliberate effort to avoid using any advanced calculus, probability, or stochastic
process concepts (although we’ve included some homework problems for students
with this advanced background). The book is therefore appropriate for undergraduate courses and for first-year graduate courses. It should also be useful to practitioners in the telecommunications industry.
What Is Unique about This Textbook?
The subject of computer networking is enormously complex, involving many
concepts, protocols, and technologies that are woven together in an intricate
manner. To cope with this scope and complexity, many computer networking texts
are often organized around the “layers” of a network architecture. With a layered
organization, students can see through the complexity of computer networking—
they learn about the distinct concepts and protocols in one part of the architecture
while seeing the big picture of how all parts fit together. From a pedagogical
perspective, our personal experience has been that such a layered approach
Preface
indeed works well. Nevertheless, we have found that the traditional approach of
teaching—bottom up; that is, from the physical layer towards the application
layer—is not the best approach for a modern course on computer networking.
A Top-Down Approach
Our book broke new ground 12 years ago by treating networking in a top-down
manner—that is, by beginning at the application layer and working its way down
toward the physical layer. The feedback we received from teachers and students
alike have confirmed that this top-down approach has many advantages and does
indeed work well pedagogically. First, it places emphasis on the application layer
(a “high growth area” in networking). Indeed, many of the recent revolutions in
computer networking—including the Web, peer-to-peer file sharing, and media
streaming—have taken place at the application layer. An early emphasis on applicationlayer issues differs from the approaches taken in most other texts, which have only a
small amount of material on network applications, their requirements, application-layer
paradigms (e.g., client-server and peer-to-peer), and application programming interfaces. Second, our experience as instructors (and that of many instructors who have
used this text) has been that teaching networking applications near the beginning of
the course is a powerful motivational tool. Students are thrilled to learn about how
networking applications work—applications such as e-mail and the Web, which most
students use on a daily basis. Once a student understands the applications, the student
can then understand the network services needed to support these applications. The
student can then, in turn, examine the various ways in which such services might be
provided and implemented in the lower layers. Covering applications early thus provides motivation for the remainder of the text.
Third, a top-down approach enables instructors to introduce network application development at an early stage. Students not only see how popular applications and protocols work, but also learn how easy it is to create their own
network applications and application-level protocols. With the top-down
approach, students get early exposure to the notions of socket programming, service models, and protocols—important concepts that resurface in all subsequent
layers. By providing socket programming examples in Python, we highlight the
central ideas without confusing students with complex code. Undergraduates in
electrical engineering and computer science should not have difficulty following
the Python code.
An Internet Focus
Although we dropped the phrase “Featuring the Internet” from the title of this book
with the fourth edition, this doesn’t mean that we dropped our focus on the Internet!
Indeed, nothing could be further from the case! Instead, since the Internet has
become so pervasive, we felt that any networking textbook must have a significant
ix
x
Preface
focus on the Internet, and thus this phrase was somewhat unnecessary. We continue
to use the Internet’s architecture and protocols as primary vehicles for studying fundamental computer networking concepts. Of course, we also include concepts and
protocols from other network architectures. But the spotlight is clearly on the Internet, a fact reflected in our organizing the book around the Internet’s five-layer architecture: the application, transport, network, link, and physical layers.
Another benefit of spotlighting the Internet is that most computer science and
electrical engineering students are eager to learn about the Internet and its protocols.
They know that the Internet has been a revolutionary and disruptive technology and
can see that it is profoundly changing our world. Given the enormous relevance of
the Internet, students are naturally curious about what is “under the hood.” Thus, it
is easy for an instructor to get students excited about basic principles when using the
Internet as the guiding focus.
Teaching Networking Principles
Two of the unique features of the book—its top-down approach and its focus on the
Internet—have appeared in the titles of our book. If we could have squeezed a third
phrase into the subtitle, it would have contained the word principles. The field of
networking is now mature enough that a number of fundamentally important issues
can be identified. For example, in the transport layer, the fundamental issues include
reliable communication over an unreliable network layer, connection establishment/
teardown and handshaking, congestion and flow control, and multiplexing. Two fundamentally important network-layer issues are determining “good” paths between
two routers and interconnecting a large number of heterogeneous networks. In the
link layer, a fundamental problem is sharing a multiple access channel. In network
security, techniques for providing confidentiality, authentication, and message
integrity are all based on cryptographic fundamentals. This text identifies fundamental networking issues and studies approaches towards addressing these issues. The
student learning these principles will gain knowledge with a long “shelf life”—long
after today’s network standards and protocols have become obsolete, the principles
they embody will remain important and relevant. We believe that the combination of
using the Internet to get the student’s foot in the door and then emphasizing fundamental issues and solution approaches will allow the student to quickly understand
just about any networking technology.
The Web Site
Each new copy of this textbook includes six months of access to a Companion Web site
for all book readers at http://www.pearsonhighered.com/kurose-ross, which includes:
• Interactive learning material. An important new component of the sixth edition
is the significantly expanded online and interactive learning material. The
book’s Companion Web site now contains VideoNotes—video presentations of
Preface
important topics thoughout the book done by the authors, as well as walkthroughs of solutions to problems similar to those at the end of the chapter.
We’ve also added Interactive Exercises that can create (and present solutions
for) problems similar to selected end-of-chapter problems. Since students can
generate (and view solutions for) an unlimited number of similar problem
instances, they can work until the material is truly mastered. We’ve seeded the
Web site with VideoNotes and online problems for chapters 1 through 5 and will
continue to actively add and update this material over time. As in earlier editions, the Web site contains the interactive Java applets that animate many key
networking concepts. The site also has interactive quizzes that permit students
to check their basic understanding of the subject matter. Professors can integrate
these interactive features into their lectures or use them as mini labs.
• Additional technical material. As we have added new material in each edition of
our book, we’ve had to remove coverage of some existing topics to keep the
book at manageable length. For example, to make room for the new material in
this edition, we’ve removed material on ATM networks and the RTSP protocol
for multimedia. Material that appeared in earlier editions of the text is still of
interest, and can be found on the book’s Web site.
• Programming assignments. The Web site also provides a number of detailed
programming assignments, which include building a multithreaded Web
server, building an e-mail client with a GUI interface, programming the sender
and receiver sides of a reliable data transport protocol, programming a distributed routing algorithm, and more.
• Wireshark labs. One’s understanding of network protocols can be greatly deepened by seeing them in action. The Web site provides numerous Wireshark
assignments that enable students to actually observe the sequence of messages
exchanged between two protocol entities. The Web site includes separate Wireshark labs on HTTP, DNS, TCP, UDP, IP, ICMP, Ethernet, ARP, WiFi, SSL, and
on tracing all protocols involved in satisfying a request to fetch a web page.
We’ll continue to add new labs over time.
Pedagogical Features
We have each been teaching computer networking for more than 20 years.
Together, we bring more than 50 years of teaching experience to this text, during
which time we have taught many thousands of students. We have also been active
researchers in computer networking during this time. (In fact, Jim and Keith first
met each other as master’s students in a computer networking course taught by
Mischa Schwartz in 1979 at Columbia University.) We think all this gives us a
good perspective on where networking has been and where it is likely to go in the
future. Nevertheless, we have resisted temptations to bias the material in this book
xi
xii
Preface
towards our own pet research projects. We figure you can visit our personal Web
sites if you are interested in our research. Thus, this book is about modern computer networking—it is about contemporary protocols and technologies as well as
the underlying principles behind these protocols and technologies. We also believe
that learning (and teaching!) about networking can be fun. A sense of humor, use
of analogies, and real-world examples in this book will hopefully make this material more fun.
Supplements for Instructors
We provide a complete supplements package to aid instructors in teaching this course.
This material can be accessed from Pearson’s Instructor Resource Center
(http://www.pearsonhighered.com/irc). Visit the Instructor Resource Center or send
e-mail to [email protected] for information about accessing these instructor’s
supplements.
• PowerPoint® slides. We provide PowerPoint slides for all nine chapters. The
slides have been completely updated with this sixth edition. The slides cover
each chapter in detail. They use graphics and animations (rather than relying
only on monotonous text bullets) to make the slides interesting and visually
appealing. We provide the original PowerPoint slides so you can customize them
to best suit your own teaching needs. Some of these slides have been contributed
by other instructors who have taught from our book.
• Homework solutions. We provide a solutions manual for the homework problems
in the text, programming assignments, and Wireshark labs. As noted earlier, we’ve
introduced many new homework problems in the first five chapters of the book.
Chapter Dependencies
The first chapter of this text presents a self-contained overview of computer networking. Introducing many key concepts and terminology, this chapter sets the stage
for the rest of the book. All of the other chapters directly depend on this first chapter. After completing Chapter 1, we recommend instructors cover Chapters 2
through 5 in sequence, following our top-down philosophy. Each of these five chapters leverages material from the preceding chapters. After completing the first five
chapters, the instructor has quite a bit of flexibility. There are no interdependencies
among the last four chapters, so they can be taught in any order. However, each of
the last four chapters depends on the material in the first five chapters. Many
instructors first teach the first five chapters and then teach one of the last four chapters for “dessert.”
Preface
One Final Note: We’d Love to Hear from You
We encourage students and instructors to e-mail us with any comments they might
have about our book. It’s been wonderful for us to hear from so many instructors
and students from around the world about our first four editions. We’ve incorporated
many of these suggestions into later editions of the book. We also encourage instructors
to send us new homework problems (and solutions) that would complement the
current homework problems. We’ll post these on the instructor-only portion of the
Web site. We also encourage instructors and students to create new Java applets that
illustrate the concepts and protocols in this book. If you have an applet that you
think would be appropriate for this text, please submit it to us. If the applet (including
notation and terminology) is appropriate, we’ll be happy to include it on the text’s
Web site, with an appropriate reference to the applet’s authors.
So, as the saying goes, “Keep those cards and letters coming!” Seriously,
please do continue to send us interesting URLs, point out typos, disagree with
any of our claims, and tell us what works and what doesn’t work. Tell us what
you think should or shouldn’t be included in the next edition. Send your e-mail
to [email protected] and [email protected]
Acknowledgments
Since we began writing this book in 1996, many people have given us invaluable
help and have been influential in shaping our thoughts on how to best organize and
teach a networking course. We want to say A BIG THANKS to everyone who has
helped us from the earliest first drafts of this book, up to this fifth edition. We are also
very thankful to the many hundreds of readers from around the world—students, faculty, practitioners—who have sent us thoughts and comments on earlier editions of
the book and suggestions for future editions of the book. Special thanks go out to:
Al Aho (Columbia University)
Hisham Al-Mubaid (University of Houston-Clear Lake)
Pratima Akkunoor (Arizona State University)
Paul Amer (University of Delaware)
Shamiul Azom (Arizona State University)
Lichun Bao (University of California at Irvine)
Paul Barford (University of Wisconsin)
Bobby Bhattacharjee (University of Maryland)
Steven Bellovin (Columbia University)
Pravin Bhagwat (Wibhu)
Supratik Bhattacharyya (previously at Sprint)
Ernst Biersack (Eurécom Institute)
xiii
xiv
Preface
Shahid Bokhari (University of Engineering & Technology, Lahore)
Jean Bolot (Technicolor Research)
Daniel Brushteyn (former University of Pennsylvania student)
Ken Calvert (University of Kentucky)
Evandro Cantu (Federal University of Santa Catarina)
Jeff Case (SNMP Research International)
Jeff Chaltas (Sprint)
Vinton Cerf (Google)
Byung Kyu Choi (Michigan Technological University)
Bram Cohen (BitTorrent, Inc.)
Constantine Coutras (Pace University)
John Daigle (University of Mississippi)
Edmundo A. de Souza e Silva (Federal University of Rio de Janeiro)
Philippe Decuetos (Eurécom Institute)
Christophe Diot (Technicolor Research)
Prithula Dhunghel (Akamai)
Deborah Estrin (University of California, Los Angeles)
Michalis Faloutsos (University of California at Riverside)
Wu-chi Feng (Oregon Graduate Institute)
Sally Floyd (ICIR, University of California at Berkeley)
Paul Francis (Max Planck Institute)
Lixin Gao (University of Massachusetts)
JJ Garcia-Luna-Aceves (University of California at Santa Cruz)
Mario Gerla (University of California at Los Angeles)
David Goodman (NYU-Poly)
Yang Guo (Alcatel/Lucent Bell Labs)
Tim Griffin (Cambridge University)
Max Hailperin (Gustavus Adolphus College)
Bruce Harvey (Florida A&M University, Florida State University)
Carl Hauser (Washington State University)
Rachelle Heller (George Washington University)
Phillipp Hoschka (INRIA/W3C)
Wen Hsin (Park University)
Albert Huang (former University of Pennsylvania student)
Cheng Huang (Microsoft Research)
Esther A. Hughes (Virginia Commonwealth University)
Van Jacobson (Xerox PARC)
Pinak Jain (former NYU-Poly student)
Jobin James (University of California at Riverside)
Sugih Jamin (University of Michigan)
Shivkumar Kalyanaraman (IBM Research, India)
Jussi Kangasharju (University of Helsinki)
Sneha Kasera (University of Utah)
Parviz Kermani (formerly of IBM Research)
Preface
Hyojin Kim (former University of Pennsylvania student)
Leonard Kleinrock (University of California at Los Angeles)
David Kotz (Dartmouth College)
Beshan Kulapala (Arizona State University)
Rakesh Kumar (Bloomberg)
Miguel A. Labrador (University of South Florida)
Simon Lam (University of Texas)
Steve Lai (Ohio State University)
Tom LaPorta (Penn State University)
Tim-Berners Lee (World Wide Web Consortium)
Arnaud Legout (INRIA)
Lee Leitner (Drexel University)
Brian Levine (University of Massachusetts)
Chunchun Li (former NYU-Poly student)
Yong Liu (NYU-Poly)
William Liang (former University of Pennsylvania student)
Willis Marti (Texas A&M University)
Nick McKeown (Stanford University)
Josh McKinzie (Park University)
Deep Medhi (University of Missouri, Kansas City)
Bob Metcalfe (International Data Group)
Sue Moon (KAIST)
Jenni Moyer (Comcast)
Erich Nahum (IBM Research)
Christos Papadopoulos (Colorado Sate University)
Craig Partridge (BBN Technologies)
Radia Perlman (Intel)
Jitendra Padhye (Microsoft Research)
Vern Paxson (University of California at Berkeley)
Kevin Phillips (Sprint)
George Polyzos (Athens University of Economics and Business)
Sriram Rajagopalan (Arizona State University)
Ramachandran Ramjee (Microsoft Research)
Ken Reek (Rochester Institute of Technology)
Martin Reisslein (Arizona State University)
Jennifer Rexford (Princeton University)
Leon Reznik (Rochester Institute of Technology)
Pablo Rodrigez (Telefonica)
Sumit Roy (University of Washington)
Avi Rubin (Johns Hopkins University)
Dan Rubenstein (Columbia University)
Douglas Salane (John Jay College)
Despina Saparilla (Cisco Systems)
John Schanz (Comcast)
xv
xvi
Preface
Henning Schulzrinne (Columbia University)
Mischa Schwartz (Columbia University)
Ardash Sethi (University of Delaware)
Harish Sethu (Drexel University)
K. Sam Shanmugan (University of Kansas)
Prashant Shenoy (University of Massachusetts)
Clay Shields (Georgetown University)
Subin Shrestra (University of Pennsylvania)
Bojie Shu (former NYU-Poly student)
Mihail L. Sichitiu (NC State University)
Peter Steenkiste (Carnegie Mellon University)
Tatsuya Suda (University of California at Irvine)
Kin Sun Tam (State University of New York at Albany)
Don Towsley (University of Massachusetts)
David Turner (California State University, San Bernardino)
Nitin Vaidya (University of Illinois)
Michele Weigle (Clemson University)
David Wetherall (University of Washington)
Ira Winston (University of Pennsylvania)
Di Wu (Sun Yat-sen University)
Shirley Wynn (NYU-Poly)
Raj Yavatkar (Intel)
Yechiam Yemini (Columbia University)
Ming Yu (State University of New York at Binghamton)
Ellen Zegura (Georgia Institute of Technology)
Honggang Zhang (Suffolk University)
Hui Zhang (Carnegie Mellon University)
Lixia Zhang (University of California at Los Angeles)
Meng Zhang (former NYU-Poly student)
Shuchun Zhang (former University of Pennsylvania student)
Xiaodong Zhang (Ohio State University)
ZhiLi Zhang (University of Minnesota)
Phil Zimmermann (independent consultant)
Cliff C. Zou (University of Central Florida)
We also want to thank the entire Addison-Wesley team—in particular, Michael Hirsch,
Marilyn Lloyd, and Emma Snider—who have done an absolutely outstanding job on
this sixth edition (and who have put up with two very finicky authors who seem congenitally unable to meet deadlines!). Thanks also to our artists, Janet Theurer and
Patrice Rossi Calkin, for their work on the beautiful figures in this book, and to Andrea
Stefanowicz and her team at PreMediaGlobal for their wonderful production work on
this edition. Finally, a most special thanks go to Michael Hirsch, our editor at AddisonWesley, and Susan Hartman, our former editor at Addison-Wesley. This book would
not be what it is (and may well not have been at all) without their graceful management, constant encouragement, nearly infinite patience, good humor, and perseverance.
Table of Contents
Chapter 1 Computer Networks and the Internet
1.1
What Is the Internet?
1.1.1 A Nuts-and-Bolts Description
1.1.2 A Services Description
1.1.3 What Is a Protocol?
1.2
The Network Edge
1.2.1 Access Networks
1.2.2 Physical Media
1.3
The Network Core
1.3.1 Packet Switching
1.3.2 Circuit Switching
1.3.3 A Network of Networks
1.4
Delay, Loss, and Throughput in Packet-Switched Networks
1.4.1 Overview of Delay in Packet-Switched Networks
1.4.2 Queuing Delay and Packet Loss
1.4.3 End-to-End Delay
1.4.4 Throughput in Computer Networks
1.5
Protocol Layers and Their Service Models
1.5.1 Layered Architecture
1.5.2 Encapsulation
1.6
Networks Under Attack
1.7
History of Computer Networking and the Internet
1.7.1 The Development of Packet Switching: 1961–1972
1.7.2 Proprietary Networks and Internetworking: 1972–1980
1.7.3 A Proliferation of Networks: 1980–1990
1.7.4 The Internet Explosion: The 1990s
1.7.5 The New Millennium
1.8
Summary
Homework Problems and Questions
Wireshark Lab
Interview: Leonard Kleinrock
1
2
2
5
7
9
12
18
22
22
27
32
35
35
39
42
44
47
47
53
55
60
60
62
63
64
65
66
68
78
80
xvii
xviii
Table of Contents
Chapter 2 Application Layer
2.1
Principles of Network Applications
2.1.1 Network Application Architectures
2.1.2 Processes Communicating
2.1.3 Transport Services Available to Applications
2.1.4 Transport Services Provided by the Internet
2.1.5 Application-Layer Protocols
2.1.6 Network Applications Covered in This Book
2.2
The Web and HTTP
2.2.1 Overview of HTTP
2.2.2 Non-Persistent and Persistent Connections
2.2.3 HTTP Message Format
2.2.4 User-Server Interaction: Cookies
2.2.5 Web Caching
2.2.6 The Conditional GET
2.3
File Transfer: FTP
2.3.1 FTP Commands and Replies
2.4
Electronic Mail in the Internet
2.4.1 SMTP
2.4.2 Comparison with HTTP
2.4.3 Mail Message Format
2.4.4 Mail Access Protocols
2.5
DNS—The Internet’s Directory Service
2.5.1 Services Provided by DNS
2.5.2 Overview of How DNS Works
2.5.3 DNS Records and Messages
2.6
Peer-to-Peer Applications
2.6.1 P2P File Distribution
2.6.2 Distributed Hash Tables (DHTs)
2.7
Socket Programming: Creating Network Applications
2.7.1 Socket Programming with UDP
2.7.2 Socket Programming with TCP
2.8
Summary
Homework Problems and Questions
Socket Programming Assignments
Wireshark Labs: HTTP, DNS
Interview: Marc Andreessen
83
84
86
88
91
93
96
97
98
98
100
103
108
110
114
116
118
118
121
124
125
125
130
131
133
139
144
145
151
156
157
163
168
169
179
181
182
Table of Contents
Chapter 3 Transport Layer
3.1
Introduction and Transport-Layer Services
3.1.1 Relationship Between Transport and Network Layers
3.1.2 Overview of the Transport Layer in the Internet
3.2
Multiplexing and Demultiplexing
3.3
Connectionless Transport: UDP
3.3.1 UDP Segment Structure
3.3.2 UDP Checksum
3.4
Principles of Reliable Data Transfer
3.4.1 Building a Reliable Data Transfer Protocol
3.4.2 Pipelined Reliable Data Transfer Protocols
3.4.3 Go-Back-N (GBN)
3.4.4 Selective Repeat (SR)
3.5
Connection-Oriented Transport: TCP
3.5.1 The TCP Connection
3.5.2 TCP Segment Structure
3.5.3 Round-Trip Time Estimation and Timeout
3.5.4 Reliable Data Transfer
3.5.5 Flow Control
3.5.6 TCP Connection Management
3.6
Principles of Congestion Control
3.6.1 The Causes and the Costs of Congestion
3.6.2 Approaches to Congestion Control
3.6.3 Network-Assisted Congestion-Control Example:
ATM ABR Congestion Control
3.7
TCP Congestion Control
3.7.1 Fairness
3.8
Summary
Homework Problems and Questions
Programming Assignments
Wireshark Labs: TCP, UDP
Interview: Van Jacobson
Chapter 4 The Network Layer
4.1
4.2
Introduction
4.1.1 Forwarding and Routing
4.1.2 Network Service Models
Virtual Circuit and Datagram Networks
4.2.1 Virtual-Circuit Networks
4.2.2 Datagram Networks
4.2.3 Origins of VC and Datagram Networks
xix
185
186
186
189
191
198
202
202
204
206
215
218
223
230
231
233
238
242
250
252
259
259
265
266
269
279
283
285
300
301
302
305
306
308
310
313
314
317
319
xx
Table of Contents
4.3
What’s Inside a Router?
4.3.1 Input Processing
4.3.2 Switching
4.3.3 Output Processing
4.3.4 Where Does Queuing Occur?
4.3.5 The Routing Control Plane
4.4
The Internet Protocol (IP): Forwarding and Addressing in the Internet
4.4.1 Datagram Format
4.4.2 IPv4 Addressing
4.4.3 Internet Control Message Protocol (ICMP)
4.4.4 IPv6
4.4.5 A Brief Foray into IP Security
4.5
Routing Algorithms
4.5.1 The Link-State (LS) Routing Algorithm
4.5.2 The Distance-Vector (DV) Routing Algorithm
4.5.3 Hierarchical Routing
4.6
Routing in the Internet
4.6.1 Intra-AS Routing in the Internet: RIP
4.6.2 Intra-AS Routing in the Internet: OSPF
4.6.3 Inter-AS Routing: BGP
4.7
Broadcast and Multicast Routing
4.7.1 Broadcast Routing Algorithms
4.7.2 Multicast
4.8
Summary
Homework Problems and Questions
Programming Assignments
Wireshark Labs: IP, ICMP
Interview: Vinton G. Cerf
Chapter 5 The Link Layer: Links, Access Networks, and LANs
5.1
5.2
5.3
Introduction to the Link Layer
5.1.1 The Services Provided by the Link Layer
5.1.2 Where Is the Link Layer Implemented?
Error-Detection and -Correction Techniques
5.2.1 Parity Checks
5.2.2 Checksumming Methods
5.2.3 Cyclic Redundancy Check (CRC)
Multiple Access Links and Protocols
5.3.1 Channel Partitioning Protocols
5.3.2 Random Access Protocols
5.3.3 Taking-Turns Protocols
5.3.4 DOCSIS: The Link-Layer Protocol for Cable Internet Access
320
322
324
326
327
331
331
332
338
353
356
362
363
366
371
379
383
384
388
390
399
400
405
412
413
429
430
431
433
434
436
437
438
440
442
443
445
448
449
459
460
Table of Contents
5.4
Switched Local Area Networks
5.4.1 Link-Layer Addressing and ARP
5.4.2 Ethernet
5.4.3 Link-Layer Switches
5.4.4 Virtual Local Area Networks (VLANs)
5.5
Link Virtualization: A Network as a Link Layer
5.5.1 Multiprotocol Label Switching (MPLS)
5.6
Data Center Networking
5.7
Retrospective: A Day in the Life of a Web Page Request
5.7.1 Getting Started: DHCP, UDP, IP, and Ethernet
5.7.2 Still Getting Started: DNS and ARP
5.7.3 Still Getting Started: Intra-Domain Routing to the DNS Server
5.7.4 Web Client-Server Interaction: TCP and HTTP
5.8
Summary
Homework Problems and Questions
Wireshark Labs: Ethernet and ARP, DHCP
Interview: Simon S. Lam
Chapter 6 Wireless and Mobile Networks
6.1
6.2
6.3
6.4
6.5
6.6
6.7
Introduction
Wireless Links and Network Characteristics
6.2.1 CDMA
WiFi: 802.11 Wireless LANs
6.3.1 The 802.11 Architecture
6.3.2 The 802.11 MAC Protocol
6.3.3 The IEEE 802.11 Frame
6.3.4 Mobility in the Same IP Subnet
6.3.5 Advanced Features in 802.11
6.3.6 Personal Area Networks: Bluetooth and Zigbee
Cellular Internet Access
6.4.1 An Overview of Cellular Network Architecture
6.4.2 3G Cellular Data Networks: Extending the Internet to Cellular
Subscribers
6.4.3 On to 4G: LTE
Mobility Management: Principles
6.5.1 Addressing
6.5.2 Routing to a Mobile Node
Mobile IP
Managing Mobility in Cellular Networks
6.7.1 Routing Calls to a Mobile User
6.7.2 Handoffs in GSM
xxi
461
462
469
476
482
486
487
490
495
495
497
498
499
500
502
510
511
513
514
519
522
526
527
531
537
541
542
544
546
547
550
553
555
557
559
564
570
571
572
xxii
Table of Contents
6.8
Wireless and Mobility: Impact on Higher-Layer Protocols
6.9
Summary
Homework Problems and Questions
Wireshark Lab: IEEE 802.11 (WiFi)
Interview: Deborah Estrin
Chapter 7 Multimedia Networking
7.1
Multimedia Networking Applications
7.1.1 Properties of Video
7.1.2 Properties of Audio
7.1.3 Types of Multimedia Network Applications
7.2
Streaming Stored Video
7.2.1 UDP Streaming
7.2.2 HTTP Streaming
7.2.3 Adaptive Streaming and DASH
7.2.4 Content Distribution Networks
7.2.5 Case Studies: Netflix, YouTube, and Kankan
7.3
Voice-over-IP
7.3.1 Limitations of the Best-Effort IP Service
7.3.2 Removing Jitter at the Receiver for Audio
7.3.3 Recovering from Packet Loss
7.3.4 Case Study: VoIP with Skype
7.4
Protocols for Real-Time Conversational Applications
7.4.1 RTP
7.4.2 SIP
7.5
Network Support for Multimedia
7.5.1 Dimensioning Best-Effort Networks
7.5.2 Providing Multiple Classes of Service
7.5.3 Diffserv
7.5.4 Per-Connection Quality-of-Service (QoS) Guarantees:
Resource Reservation and Call Admission
7.6
Summary
Homework Problems and Questions
Programming Assignment
Interview: Henning Schulzrinne
Chapter 8 Security in Computer Networks
8.1
8.2
What Is Network Security?
Principles of Cryptography
8.2.1 Symmetric Key Cryptography
8.2.2 Public Key Encryption
575
578
578
583
584
587
588
588
590
591
593
595
596
600
602
608
612
612
614
617
620
623
624
627
632
634
636
648
652
655
656
666
668
671
672
675
676
683
Table of Contents
8.3
Message Integrity and Digital Signatures
8.3.1 Cryptographic Hash Functions
8.3.2 Message Authentication Code
8.3.3 Digital Signatures
8.4
End-Point Authentication
8.4.1 Authentication Protocol ap1.0
8.4.2 Authentication Protocol ap2.0
8.4.3 Authentication Protocol ap3.0
8.4.4 Authentication Protocol ap3.1
8.4.5 Authentication Protocol ap4.0
8.5
Securing E-Mail
8.5.1 Secure E-Mail
8.5.2 PGP
8.6
Securing TCP Connections: SSL
8.6.1 The Big Picture
8.6.2 A More Complete Picture
8.7
Network-Layer Security: IPsec and Virtual Private Networks
8.7.1 IPsec and Virtual Private Networks (VPNs)
8.7.2 The AH and ESP Protocols
8.7.3 Security Associations
8.7.4 The IPsec Datagram
8.7.5 IKE: Key Management in IPsec
8.8
Securing Wireless LANs
8.8.1 Wired Equivalent Privacy (WEP)
8.8.2 IEEE 802.11i
8.9
Operational Security: Firewalls and Intrusion Detection Systems
8.9.1 Firewalls
8.9.2 Intrusion Detection Systems
8.10 Summary
Homework Problems and Questions
Wireshark Lab: SSL
IPsec Lab
Interview: Steven M. Bellovin
Chapter 9 Network Management
9.1
9.2
9.3
What Is Network Management?
The Infrastructure for Network Management
The Internet-Standard Management Framework
9.3.1 Structure of Management Information: SMI
9.3.2 Management Information Base: MIB
xxiii
688
689
691
693
700
700
701
702
703
703
705
706
710
711
713
716
718
718
720
720
721
725
726
726
728
731
731
739
742
744
752
752
753
755
756
760
764
766
770
xxiv
Table of Contents
9.3.3 SNMP Protocol Operations and Transport Mappings
9.3.4 Security and Administration
9.4
ASN.1
9.5
Conclusion
Homework Problems and Questions
Interview: Jennifer Rexford
References
Index
772
775
778
783
783
786
789
823
COMPUTER
NETWORKING
SIXTH EDITION
A Top-Down Approach
This page intentionally left blank
CHAPTER
1
Computer
Networks and
the Internet
Today’s Internet is arguably the largest engineered system ever created by mankind,
with hundreds of millions of connected computers, communication links, and
switches; with billions of users who connect via laptops, tablets, and smartphones;
and with an array of new Internet-connected devices such as sensors, Web cams,
game consoles, picture frames, and even washing machines. Given that the Internet
is so large and has so many diverse components and uses, is there any hope of
understanding how it works? Are there guiding principles and structure that can provide a foundation for understanding such an amazingly large and complex system?
And if so, is it possible that it actually could be both interesting and fun to learn
about computer networks? Fortunately, the answers to all of these questions is a
resounding YES! Indeed, it’s our aim in this book to provide you with a modern
introduction to the dynamic field of computer networking, giving you the principles
and practical insights you’ll need to understand not only today’s networks, but
tomorrow’s as well.
This first chapter presents a broad overview of computer networking and the
Internet. Our goal here is to paint a broad picture and set the context for the rest of
this book, to see the forest through the trees. We’ll cover a lot of ground in this introductory chapter and discuss a lot of the pieces of a computer network, without losing sight of the big picture.
1
2
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
We’ll structure our overview of computer networks in this chapter as follows.
After introducing some basic terminology and concepts, we’ll first examine the
basic hardware and software components that make up a network. We’ll begin at
the network’s edge and look at the end systems and network applications running
in the network. We’ll then explore the core of a computer network, examining the
links and the switches that transport data, as well as the access networks and physical media that connect end systems to the network core. We’ll learn that the Internet is a network of networks, and we’ll learn how these networks connect with
each other.
After having completed this overview of the edge and core of a computer network, we’ll take the broader and more abstract view in the second half of this chapter. We’ll examine delay, loss, and throughput of data in a computer network and
provide simple quantitative models for end-to-end throughput and delay: models
that take into account transmission, propagation, and queuing delays. We’ll then
introduce some of the key architectural principles in computer networking, namely,
protocol layering and service models. We’ll also learn that computer networks are
vulnerable to many different types of attacks; we’ll survey some of these attacks and
consider how computer networks can be made more secure. Finally, we’ll close this
chapter with a brief history of computer networking.
1.1 What Is the Internet?
In this book, we’ll use the public Internet, a specific computer network, as our principal vehicle for discussing computer networks and their protocols. But what is the
Internet? There are a couple of ways to answer this question. First, we can describe
the nuts and bolts of the Internet, that is, the basic hardware and software components
that make up the Internet. Second, we can describe the Internet in terms of a networking infrastructure that provides services to distributed applications. Let’s begin
with the nuts-and-bolts description, using Figure 1.1 to illustrate our discussion.
1.1.1 A Nuts-and-Bolts Description
The Internet is a computer network that interconnects hundreds of millions of computing devices throughout the world. Not too long ago, these computing devices were
primarily traditional desktop PCs, Linux workstations, and so-called servers that store
and transmit information such as Web pages and e-mail messages. Increasingly,
however, nontraditional Internet end systems such as laptops, smartphones, tablets,
TVs, gaming consoles, Web cams, automobiles, environmental sensing devices,
picture frames, and home electrical and security systems are being connected to the
Internet. Indeed, the term computer network is beginning to sound a bit dated, given
the many nontraditional devices that are being hooked up to the Internet. In Internet jargon, all of these devices are called hosts or end systems. As of July 2011, there were
1.1
•
WHAT IS THE INTERNET?
National or
Global ISP
Mobile Network
Local or
Regional ISP
Home Network
Enterprise Network
Key:
Host
Server
(= end system)
Figure 1.1
Mobile
Router
Link-Layer
switch
Some pieces of the Internet
Modem
Base
station
Smartphone Cell phone
tower
3
4
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
nearly 850 million end systems attached to the Internet [ISC 2012], not counting
smartphones, laptops, and other devices that are only intermittently connected to the
Internet. Overall, more there are an estimated 2 billion Internet users [ITU 2011].
End systems are connected together by a network of communication links and
packet switches. We’ll see in Section 1.2 that there are many types of communication links, which are made up of different types of physical media, including coaxial
cable, copper wire, optical fiber, and radio spectrum. Different links can transmit
data at different rates, with the transmission rate of a link measured in bits/second.
When one end system has data to send to another end system, the sending end system segments the data and adds header bytes to each segment. The resulting packages of information, known as packets in the jargon of computer networks, are then
sent through the network to the destination end system, where they are reassembled
into the original data.
A packet switch takes a packet arriving on one of its incoming communication
links and forwards that packet on one of its outgoing communication links. Packet
switches come in many shapes and flavors, but the two most prominent types in
today’s Internet are routers and link-layer switches. Both types of switches forward packets toward their ultimate destinations. Link-layer switches are typically
used in access networks, while routers are typically used in the network core. The
sequence of communication links and packet switches traversed by a packet from
the sending end system to the receiving end system is known as a route or path
through the network. The exact amount of traffic being carried in the Internet is
difficult to estimate but Cisco [Cisco VNI 2011] estimates global Internet traffic will
be nearly 40 exabytes per month in 2012.
Packet-switched networks (which transport packets) are in many ways similar to transportation networks of highways, roads, and intersections (which transport vehicles). Consider, for example, a factory that needs to move a large
amount of cargo to some destination warehouse located thousands of kilometers
away. At the factory, the cargo is segmented and loaded into a fleet of trucks.
Each of the trucks then independently travels through the network of highways,
roads, and intersections to the destination warehouse. At the destination warehouse, the cargo is unloaded and grouped with the rest of the cargo arriving from
the same shipment. Thus, in many ways, packets are analogous to trucks, communication links are analogous to highways and roads, packet switches are analogous to intersections, and end systems are analogous to buildings. Just as a truck
takes a path through the transportation network, a packet takes a path through a
computer network.
End systems access the Internet through Internet Service Providers (ISPs),
including residential ISPs such as local cable or telephone companies; corporate
ISPs; university ISPs; and ISPs that provide WiFi access in airports, hotels, coffee
shops, and other public places. Each ISP is in itself a network of packet switches
and communication links. ISPs provide a variety of types of network access to the
end systems, including residential broadband access such as cable modem or DSL,
1.1
•
WHAT IS THE INTERNET?
high-speed local area network access, wireless access, and 56 kbps dial-up modem
access. ISPs also provide Internet access to content providers, connecting Web
sites directly to the Internet. The Internet is all about connecting end systems to
each other, so the ISPs that provide access to end systems must also be interconnected. These lower-tier ISPs are interconnected through national and international upper-tier ISPs such as Level 3 Communications, AT&T, Sprint, and NTT.
An upper-tier ISP consists of high-speed routers interconnected with high-speed
fiber-optic links. Each ISP network, whether upper-tier or lower-tier, is managed
independently, runs the IP protocol (see below), and conforms to certain naming
and address conventions. We’ll examine ISPs and their interconnection more
closely in Section 1.3.
End systems, packet switches, and other pieces of the Internet run protocols
that control the sending and receiving of information within the Internet. The
Transmission Control Protocol (TCP) and the Internet Protocol (IP) are two of
the most important protocols in the Internet. The IP protocol specifies the format of
the packets that are sent and received among routers and end systems. The Internet’s
principal protocols are collectively known as TCP/IP. We’ll begin looking into protocols in this introductory chapter. But that’s just a start—much of this book is concerned with computer network protocols!
Given the importance of protocols to the Internet, it’s important that everyone
agree on what each and every protocol does, so that people can create systems and
products that interoperate. This is where standards come into play. Internet standards are developed by the Internet Engineering Task Force (IETF)[IETF 2012].
The IETF standards documents are called requests for comments (RFCs). RFCs
started out as general requests for comments (hence the name) to resolve network
and protocol design problems that faced the precursor to the Internet [Allman 2011].
RFCs tend to be quite technical and detailed. They define protocols such as TCP, IP,
HTTP (for the Web), and SMTP (for e-mail). There are currently more than 6,000
RFCs. Other bodies also specify standards for network components, most notably
for network links. The IEEE 802 LAN/MAN Standards Committee [IEEE 802
2012], for example, specifies the Ethernet and wireless WiFi standards.
1.1.2 A Services Description
Our discussion above has identified many of the pieces that make up the Internet.
But we can also describe the Internet from an entirely different angle—namely, as
an infrastructure that provides services to applications. These applications
include electronic mail, Web surfing, social networks, instant messaging, Voiceover-IP (VoIP), video streaming, distributed games, peer-to-peer (P2P) file sharing, television over the Internet, remote login, and much, much more. The
applications are said to be distributed applications, since they involve multiple
end systems that exchange data with each other. Importantly, Internet applications
5
6
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
run on end systems—they do not run in the packet switches in the network core.
Although packet switches facilitate the exchange of data among end systems, they
are not concerned with the application that is the source or sink of data.
Let’s explore a little more what we mean by an infrastructure that provides
services to applications. To this end, suppose you have an exciting new idea for a
distributed Internet application, one that may greatly benefit humanity or one that
may simply make you rich and famous. How might you go about transforming
this idea into an actual Internet application? Because applications run on end systems, you are going to need to write programs that run on the end systems. You
might, for example, write your programs in Java, C, or Python. Now, because you
are developing a distributed Internet application, the programs running on the
different end systems will need to send data to each other. And here we get to a
central issue—one that leads to the alternative way of describing the Internet as a
platform for applications. How does one program running on one end system
instruct the Internet to deliver data to another program running on another end
system?
End systems attached to the Internet provide an Application Programming
Interface (API) that specifies how a program running on one end system asks
the Internet infrastructure to deliver data to a specific destination program running on another end system. This Internet API is a set of rules that the sending
program must follow so that the Internet can deliver the data to the destination
program. We’ll discuss the Internet API in detail in Chapter 2. For now, let’s
draw upon a simple analogy, one that we will frequently use in this book. Suppose Alice wants to send a letter to Bob using the postal service. Alice, of course,
can’t just write the letter (the data) and drop the letter out her window. Instead,
the postal service requires that Alice put the letter in an envelope; write Bob’s
full name, address, and zip code in the center of the envelope; seal the envelope;
put a stamp in the upper-right-hand corner of the envelope; and finally, drop the
envelope into an official postal service mailbox. Thus, the postal service has its
own “postal service API,” or set of rules, that Alice must follow to have the
postal service deliver her letter to Bob. In a similar manner, the Internet has an
API that the program sending data must follow to have the Internet deliver the
data to the program that will receive the data.
The postal service, of course, provides more than one service to its customers.
It provides express delivery, reception confirmation, ordinary use, and many more
services. In a similar manner, the Internet provides multiple services to its applications. When you develop an Internet application, you too must choose one of the
Internet’s services for your application. We’ll describe the Internet’s services in
Chapter 2.
We have just given two descriptions of the Internet; one in terms of its hardware
and software components, the other in terms of an infrastructure for providing
services to distributed applications. But perhaps you are still confused as to what the
1.1
•
WHAT IS THE INTERNET?
Internet is. What are packet switching and TCP/IP? What are routers? What kinds of
communication links are present in the Internet? What is a distributed application?
How can a toaster or a weather sensor be attached to the Internet? If you feel a bit
overwhelmed by all of this now, don’t worry—the purpose of this book is to introduce you to both the nuts and bolts of the Internet and the principles that govern how
and why it works. We’ll explain these important terms and questions in the following sections and chapters.
1.1.3 What Is a Protocol?
Now that we’ve got a bit of a feel for what the Internet is, let’s consider another
important buzzword in computer networking: protocol. What is a protocol? What
does a protocol do?
A Human Analogy
It is probably easiest to understand the notion of a computer network protocol by
first considering some human analogies, since we humans execute protocols all of
the time. Consider what you do when you want to ask someone for the time of day.
A typical exchange is shown in Figure 1.2. Human protocol (or good manners, at
least) dictates that one first offer a greeting (the first “Hi” in Figure 1.2) to initiate
communication with someone else. The typical response to a “Hi” is a returned
“Hi” message. Implicitly, one then takes a cordial “Hi” response as an indication
that one can proceed and ask for the time of day. A different response to the initial
“Hi” (such as “Don’t bother me!” or “I don’t speak English,” or some unprintable
reply) might indicate an unwillingness or inability to communicate. In this case,
the human protocol would be not to ask for the time of day. Sometimes one gets no
response at all to a question, in which case one typically gives up asking that person for the time. Note that in our human protocol, there are specific messages we
send, and specific actions we take in response to the received reply messages or
other events (such as no reply within some given amount of time). Clearly, transmitted and received messages, and actions taken when these messages are sent or
received or other events occur, play a central role in a human protocol. If people
run different protocols (for example, if one person has manners but the other does
not, or if one understands the concept of time and the other does not) the protocols
do not interoperate and no useful work can be accomplished. The same is true in
networking—it takes two (or more) communicating entities running the same protocol in order to accomplish a task.
Let’s consider a second human analogy. Suppose you’re in a college class (a
computer networking class, for example!). The teacher is droning on about protocols and you’re confused. The teacher stops to ask, “Are there any questions?” (a
7
8
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
TCP
Hi
conn
Hi
TCP
GET
Got
the
time
?
2:00
p:/
Time
n re
ctio
nne
/ww
w.a
que
n re
wl.
<file
Time
Figure 1.2
htt
co
ectio
com
st
ply
/ku
ros
e-r
oss
>
Time
Time
A human protocol and a computer network protocol
message that is transmitted to, and received by, all students who are not sleeping).
You raise your hand (transmitting an implicit message to the teacher). Your teacher
acknowledges you with a smile, saying “Yes . . .” (a transmitted message encouraging you to ask your question—teachers love to be asked questions), and you then ask
your question (that is, transmit your message to your teacher). Your teacher hears
your question (receives your question message) and answers (transmits a reply to
you). Once again, we see that the transmission and receipt of messages, and a set of
conventional actions taken when these messages are sent and received, are at the
heart of this question-and-answer protocol.
Network Protocols
A network protocol is similar to a human protocol, except that the entities exchanging messages and taking actions are hardware or software components of some
device (for example, computer, smartphone, tablet, router, or other network-capable
1.2
•
THE NETWORK EDGE
device). All activity in the Internet that involves two or more communicating remote
entities is governed by a protocol. For example, hardware-implemented protocols in
two physically connected computers control the flow of bits on the “wire” between
the two network interface cards; congestion-control protocols in end systems control the rate at which packets are transmitted between sender and receiver; protocols
in routers determine a packet’s path from source to destination. Protocols are running everywhere in the Internet, and consequently much of this book is about computer network protocols.
As an example of a computer network protocol with which you are probably
familiar, consider what happens when you make a request to a Web server, that is,
when you type the URL of a Web page into your Web browser. The scenario is illustrated in the right half of Figure 1.2. First, your computer will send a connection
request message to the Web server and wait for a reply. The Web server will eventually receive your connection request message and return a connection reply message. Knowing that it is now OK to request the Web document, your computer then
sends the name of the Web page it wants to fetch from that Web server in a GET
message. Finally, the Web server returns the Web page (file) to your computer.
Given the human and networking examples above, the exchange of messages
and the actions taken when these messages are sent and received are the key defining elements of a protocol:
A protocol defines the format and the order of messages exchanged between
two or more communicating entities, as well as the actions taken on the transmission and/or receipt of a message or other event.
The Internet, and computer networks in general, make extensive use of protocols. Different protocols are used to accomplish different communication tasks. As
you read through this book, you will learn that some protocols are simple and
straightforward, while others are complex and intellectually deep. Mastering the
field of computer networking is equivalent to understanding the what, why, and how
of networking protocols.
1.2 The Network Edge
In the previous section we presented a high-level overview of the Internet and networking protocols. We are now going to delve a bit more deeply into the components of a computer network (and the Internet, in particular). We begin in this
section at the edge of a network and look at the components with which we are most
familiar—namely, the computers, smartphones and other devices that we use on a
daily basis. In the next section we’ll move from the network edge to the network
core and examine switching and routing in computer networks.
9
10
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
CASE HISTORY
A DIZZYING ARRAY OF INTERNET END SYSTEMS
Not too long ago, the end-system devices connected to the Internet were primarily
traditional computers such as desktop machines and powerful servers. Beginning in
the late 1990s and continuing today, a wide range of interesting devices are being
connected to the Internet, leveraging their ability to send and receive digital data.
Given the Internet’s ubiquity, its well-defined (standardized) protocols, and the
availability of Internet-ready commodity hardware, it’s natural to use Internet technology to network these devices together and to Internet-connected servers.
Many of these devices are based in the home—video game consoles (e.g.,
Microsoft’s Xbox), Internet-ready televisions, digital picture frames that download
and display digital pictures, washing machines, refrigerators, and even a toaster
that downloads meteorological information and burns an image of the day’s forecast (e.g., mixed clouds and sun) on your morning toast [BBC 2001]. IP-enabled
phones with GPS capabilities put location-dependent services (maps, information
about nearby services or people) at your fingertips. Networked sensors embedded
into the physical environment allow monitoring of buildings, bridges, seismic activity, wildlife habitats, river estuaries, and the weather. Biomedical devices can be
embedded and networked in a body-area network. With so many diverse devices
being networked together, the Internet is indeed becoming an “Internet of things”
[ITU 2005b].
Recall from the previous section that in computer networking jargon, the computers and other devices connected to the Internet are often referred to as end systems. They are referred to as end systems because they sit at the edge of the Internet,
as shown in Figure 1.3. The Internet’s end systems include desktop computers (e.g.,
desktop PCs, Macs, and Linux boxes), servers (e.g., Web and e-mail servers), and
mobile computers (e.g., laptops, smartphones, and tablets). Furthermore, an increasing number of non-traditional devices are being attached to the Internet as end systems (see sidebar).
End systems are also referred to as hosts because they host (that is, run) application programs such as a Web browser program, a Web server program, an e-mail
client program, or an e-mail server program. Throughout this book we will use the
terms hosts and end systems interchangeably; that is, host = end system. Hosts are
sometimes further divided into two categories: clients and servers. Informally,
clients tend to be desktop and mobile PCs, smartphones, and so on, whereas servers
tend to be more powerful machines that store and distribute Web pages, stream
video, relay e-mail, and so on. Today, most of the servers from which we receive
1.2
•
THE NETWORK EDGE
National or
Global ISP
Mobile Network
Local or
Regional ISP
Home Network
Enterprise Network
Figure 1.3
End-system interaction
search results, e-mail, Web pages, and videos reside in large data centers. For
example, Google has 30–50 data centers, with many having more than one hundred
thousand servers.
11
12
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
1.2.1 Access Networks
Having considered the applications and end systems at the “edge of the network,”
let’s next consider the access network—the network that physically connects an end
system to the first router (also known as the “edge router”) on a path from the end
system to any other distant end system. Figure 1.4 shows several types of access
National or
Global ISP
Mobile Network
Local or
Regional ISP
Home Network
Enterprise Network
Figure 1.4
Access networks
1.2
•
THE NETWORK EDGE
networks with thick, shaded lines, and the settings (home, enterprise, and wide-area
mobile wireless) in which they are used.
Home Access: DSL, Cable, FTTH, Dial-Up, and Satellite
In developed countries today, more than 65 percent of the households have Internet
access, with Korea, Netherlands, Finland, and Sweden leading the way with more than
80 percent of households having Internet access, almost all via a high-speed broadband
connection [ITU 2011]. Finland and Spain have recently declared high-speed Internet
access to be a “legal right.” Given this intense interest in home access, let’s begin our
overview of access networks by considering how homes connect to the Internet.
Today, the two most prevalent types of broadband residential access are digital
subscriber line (DSL) and cable. A residence typically obtains DSL Internet access
from the same local telephone company (telco) that provides its wired local phone
access. Thus, when DSL is used, a customer’s telco is also its ISP. As shown in
Figure 1.5, each customer’s DSL modem uses the existing telephone line (twistedpair copper wire, which we’ll discuss in Section 1.2.2) to exchange data with a digital subscriber line access multiplexer (DSLAM) located in the telco’s local central
office (CO). The home’s DSL modem takes digital data and translates it to highfrequency tones for transmission over telephone wires to the CO; the analog signals
from many such houses are translated back into digital format at the DSLAM.
The residential telephone line carries both data and traditional telephone signals simultaneously, which are encoded at different frequencies:
• A high-speed downstream channel, in the 50 kHz to 1 MHz band
• A medium-speed upstream channel, in the 4 kHz to 50 kHz band
• An ordinary two-way telephone channel, in the 0 to 4 kHz band
This approach makes the single DSL link appear as if there were three separate
links, so that a telephone call and an Internet connection can share the DSL link at
the same time. (We’ll describe this technique of frequency-division multiplexing in
Home
phone
Internet
Existing phone line:
0-4KHz phone; 4-50KHz
upstream data; 50KHz–
1MHz downstream data
Splitter
DSL
modem
Home PC
Figure 1.5
DSL Internet access
Central
office
DSLAM
Telephone
network
13
14
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
Section 1.3.1). On the customer side, a splitter separates the data and telephone signals arriving to the home and forwards the data signal to the DSL modem. On the
telco side, in the CO, the DSLAM separates the data and phone signals and sends
the data into the Internet. Hundreds or even thousands of households connect to a
single DSLAM [Dischinger 2007].
The DSL standards define transmission rates of 12 Mbps downstream and
1.8 Mbps upstream [ITU 1999], and 24 Mbps downstream and 2.5 Mbps upstream
[ITU 2003]. Because the downstream and upstream rates are different, the access is
said to be asymmetric. The actual downstream and upstream transmission rates
achieved may be less than the rates noted above, as the DSL provider may purposefully limit a residential rate when tiered service (different rates, available at different prices) are offered, or because the maximum rate can be limited by the distance
between the home and the CO, the gauge of the twisted-pair line and the degree of
electrical interference. Engineers have expressly designed DSL for short distances
between the home and the CO; generally, if the residence is not located within 5 to 10
miles of the CO, the residence must resort to an alternative form of Internet access.
While DSL makes use of the telco’s existing local telephone infrastructure,
cable Internet access makes use of the cable television company’s existing cable
television infrastructure. A residence obtains cable Internet access from the same
company that provides its cable television. As illustrated in Figure 1.6, fiber optics
connect the cable head end to neighborhood-level junctions, from which traditional coaxial cable is then used to reach individual houses and apartments. Each
neighborhood junction typically supports 500 to 5,000 homes. Because both fiber
and coaxial cable are employed in this system, it is often referred to as hybrid
fiber coax (HFC).
Coaxial cable
Hundreds
of homes
Fiber
node
Fiber
cable
Internet
CMTS
Hundreds
of homes
Fiber
node
Figure 1.6
Cable head end
A hybrid fiber-coaxial access network
1.2
•
THE NETWORK EDGE
Cable internet access requires special modems, called cable modems. As with a
DSL modem, the cable modem is typically an external device and connects to the
home PC through an Ethernet port. (We will discuss Ethernet in great detail in
Chapter 5.) At the cable head end, the cable modem termination system (CMTS)
serves a similar function as the DSL network’s DSLAM—turning the analog signal
sent from the cable modems in many downstream homes back into digital format.
Cable modems divide the HFC network into two channels, a downstream and an
upstream channel. As with DSL, access is typically asymmetric, with the downstream channel typically allocated a higher transmission rate than the upstream
channel. The DOCSIS 2.0 standard defines downstream rates up to 42.8 Mbps and
upstream rates of up to 30.7 Mbps. As in the case of DSL networks, the maximum
achievable rate may not be realized due to lower contracted data rates or media
impairments.
One important characteristic of cable Internet access is that it is a shared
broadcast medium. In particular, every packet sent by the head end travels downstream on every link to every home and every packet sent by a home travels on the
upstream channel to the head end. For this reason, if several users are simultaneously downloading a video file on the downstream channel, the actual rate at which
each user receives its video file will be significantly lower than the aggregate cable
downstream rate. On the other hand, if there are only a few active users and they
are all Web surfing, then each of the users may actually receive Web pages at the
full cable downstream rate, because the users will rarely request a Web page at
exactly the same time. Because the upstream channel is also shared, a distributed
multiple access protocol is needed to coordinate transmissions and avoid collisions.
(We’ll discuss this collision issue in some detail in Chapter 5.)
Although DSL and cable networks currently represent more than 90 percent of
residential broadband access in the United States, an up-and-coming technology that
promises even higher speeds is the deployment of fiber to the home (FTTH)
[FTTH Council 2011a]. As the name suggests, the FTTH concept is simple—
provide an optical fiber path from the CO directly to the home. In the United States,
Verizon has been particularly aggressive with FTTH with its FIOS service [Verizon
FIOS 2012].
There are several competing technologies for optical distribution from the
CO to the homes. The simplest optical distribution network is called direct fiber,
with one fiber leaving the CO for each home. More commonly, each fiber leaving the central office is actually shared by many homes; it is not until the fiber
gets relatively close to the homes that it is split into individual customer-specific
fibers. There are two competing optical-distribution network architectures that
perform this splitting: active optical networks (AONs) and passive optical networks (PONs). AON is essentially switched Ethernet, which is discussed in
Chapter 5.
Here, we briefly discuss PON, which is used in Verizon’s FIOS service.
Figure 1.7 shows FTTH using the PON distribution architecture. Each home has
15
16
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
Internet
Central office
ONT
Optical
splitter
ONT
OLT
ONT
Figure 1.7
Optical
fibers
FTTH Internet access
an optical network terminator (ONT), which is connected by dedicated optical
fiber to a neighborhood splitter. The splitter combines a number of homes (typically
less than 100) onto a single, shared optical fiber, which connects to an optical line
terminator (OLT) in the telco’s CO. The OLT, providing conversion between optical
and electrical signals, connects to the Internet via a telco router. In the home, users
connect a home router (typically a wireless router) to the ONT and access the Internet via this home router. In the PON architecture, all packets sent from OLT to the
splitter are replicated at the splitter (similar to a cable head end).
FTTH can potentially provide Internet access rates in the gigabits per second
range. However, most FTTH ISPs provide different rate offerings, with the higher
rates naturally costing more money. The average downstream speed of US FTTH
customers was approximately 20 Mbps in 2011 (compared with 13 Mbps for cable
access networks and less than 5 Mbps for DSL) [FTTH Council 2011b].
Two other access network technologies are also used to provide Internet access
to the home. In locations where DSL, cable, and FTTH are not available (e.g., in
some rural settings), a satellite link can be used to connect a residence to the Internet at speeds of more than 1 Mbps; StarBand and HughesNet are two such satellite
access providers. Dial-up access over traditional phone lines is based on the same
model as DSL—a home modem connects over a phone line to a modem in the ISP.
Compared with DSL and other broadband access networks, dial-up access is excruciatingly slow at 56 kbps.
Access in the Enterprise (and the Home): Ethernet and WiFi
On corporate and university campuses, and increasingly in home settings, a local area
network (LAN) is used to connect an end system to the edge router. Although there
are many types of LAN technologies, Ethernet is by far the most prevalent access
technology in corporate, university, and home networks. As shown in Figure 1.8,
Ethernet users use twisted-pair copper wire to connect to an Ethernet switch, a
1.2
•
THE NETWORK EDGE
100 Mbps
100 Mbps
Ethernet
switch
Institutional
router
To Institution’s
ISP
100 Mbps
Server
Figure 1.8
Ethernet Internet access
technology discussed in detail in Chapter 5. The Ethernet switch, or a network of
such interconnected switches, is then in turn connected into the larger Internet. With
Ethernet access, users typically have 100 Mbps access to the Ethernet switch,
whereas servers may have 1 Gbps or even 10 Gbps access.
Increasingly, however, people are accessing the Internet wirelessly from laptops, smartphones, tablets, and other devices (see earlier sidebar on “A Dizzying
Array of Devices”). In a wireless LAN setting, wireless users transmit/receive packets to/from an access point that is connected into the enterprise’s network (most
likely including wired Ethernet), which in turn is connected to the wired Internet. A
wireless LAN user must typically be within a few tens of meters of the access point.
Wireless LAN access based on IEEE 802.11 technology, more colloquially known
as WiFi, is now just about everywhere—universities, business offices, cafes, airports, homes, and even in airplanes. In many cities, one can stand on a street corner
and be within range of ten or twenty base stations (for a browseable global map of
802.11 base stations that have been discovered and logged on a Web site by people
who take great enjoyment in doing such things, see [wigle.net 2012]). As discussed
in detail in Chapter 6, 802.11 today provides a shared transmission rate of up to
54 Mbps.
Even though Ethernet and WiFi access networks were initially deployed in enterprise (corporate, university) settings, they have recently become relatively common
components of home networks. Many homes combine broadband residential access
(that is, cable modems or DSL) with these inexpensive wireless LAN technologies to
create powerful home networks [Edwards 2011]. Figure 1.9 shows a typical home
network. This home network consists of a roaming laptop as well as a wired PC; a
base station (the wireless access point), which communicates with the wireless PC; a
cable modem, providing broadband access to the Internet; and a router, which interconnects the base station and the stationary PC with the cable modem. This network
allows household members to have broadband access to the Internet with one member roaming from the kitchen to the backyard to the bedrooms.
17
18
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
Cable
head end
Internet
House
Figure 1.9
A typical home network
Wide-Area Wireless Access: 3G and LTE
Increasingly, devices such as iPhones, BlackBerrys, and Android devices are being
used to send email, surf the Web, Tweet, and download music while on the run.
These devices employ the same wireless infrastructure used for cellular telephony
to send/receive packets through a base station that is operated by the cellular network provider. Unlike WiFi, a user need only be within a few tens of kilometers (as
opposed to a few tens of meters) of the base station.
Telecommunications companies have made enormous investments in so-called
third-generation (3G) wireless, which provides packet-switched wide-area wireless
Internet access at speeds in excess of 1 Mbps. But even higher-speed wide-area
access technologies—a fourth-generation (4G) of wide-area wireless networks—are
already being deployed. LTE ( for “Long-Term Evolution”—a candidate for Bad
Acronym of the Year Award) has its roots in 3G technology, and can potentially
achieve rates in excess of 10 Mbps. LTE downstream rates of many tens of Mbps
have been reported in commercial deployments. We’ll cover the basic principles of
wireless networks and mobility, as well as WiFi, 3G, and LTE technologies (and
more!) in Chapter 6.
1.2.2 Physical Media
In the previous subsection, we gave an overview of some of the most important
network access technologies in the Internet. As we described these technologies,
we also indicated the physical media used. For example, we said that HFC uses a
combination of fiber cable and coaxial cable. We said that DSL and Ethernet use
copper wire. And we said that mobile access networks use the radio spectrum.
1.2
•
THE NETWORK EDGE
In this subsection we provide a brief overview of these and other transmission
media that are commonly used in the Internet.
In order to define what is meant by a physical medium, let us reflect on the
brief life of a bit. Consider a bit traveling from one end system, through a series of
links and routers, to another end system. This poor bit gets kicked around and
transmitted many, many times! The source end system first transmits the bit, and
shortly thereafter the first router in the series receives the bit; the first router then
transmits the bit, and shortly thereafter the second router receives the bit; and so
on. Thus our bit, when traveling from source to destination, passes through a series
of transmitter-receiver pairs. For each transmitter-receiver pair, the bit is sent by
propagating electromagnetic waves or optical pulses across a physical medium.
The physical medium can take many shapes and forms and does not have to be of
the same type for each transmitter-receiver pair along the path. Examples of physical media include twisted-pair copper wire, coaxial cable, multimode fiber-optic
cable, terrestrial radio spectrum, and satellite radio spectrum. Physical media fall
into two categories: guided media and unguided media. With guided media, the
waves are guided along a solid medium, such as a fiber-optic cable, a twisted-pair
copper wire, or a coaxial cable. With unguided media, the waves propagate in the
atmosphere and in outer space, such as in a wireless LAN or a digital satellite
channel.
But before we get into the characteristics of the various media types, let us say
a few words about their costs. The actual cost of the physical link (copper wire,
fiber-optic cable, and so on) is often relatively minor compared with other networking costs. In particular, the labor cost associated with the installation of the physical
link can be orders of magnitude higher than the cost of the material. For this reason,
many builders install twisted pair, optical fiber, and coaxial cable in every room in a
building. Even if only one medium is initially used, there is a good chance that
another medium could be used in the near future, and so money is saved by not having to lay additional wires in the future.
Twisted-Pair Copper Wire
The least expensive and most commonly used guided transmission medium is
twisted-pair copper wire. For over a hundred years it has been used by telephone
networks. In fact, more than 99 percent of the wired connections from the telephone handset to the local telephone switch use twisted-pair copper wire. Most of
us have seen twisted pair in our homes and work environments. Twisted pair consists of two insulated copper wires, each about 1 mm thick, arranged in a regular
spiral pattern. The wires are twisted together to reduce the electrical interference
from similar pairs close by. Typically, a number of pairs are bundled together in a
cable by wrapping the pairs in a protective shield. A wire pair constitutes a single
communication link. Unshielded twisted pair (UTP) is commonly used for
19
20
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
computer networks within a building, that is, for LANs. Data rates for LANs
using twisted pair today range from 10 Mbps to 10 Gbps. The data rates that can
be achieved depend on the thickness of the wire and the distance between transmitter and receiver.
When fiber-optic technology emerged in the 1980s, many people disparaged
twisted pair because of its relatively low bit rates. Some people even felt that fiberoptic technology would completely replace twisted pair. But twisted pair did not
give up so easily. Modern twisted-pair technology, such as category 6a cable, can
achieve data rates of 10 Gbps for distances up to a hundred meters. In the end,
twisted pair has emerged as the dominant solution for high-speed LAN networking.
As discussed earlier, twisted pair is also commonly used for residential Internet
access. We saw that dial-up modem technology enables access at rates of up to 56
kbps over twisted pair. We also saw that DSL (digital subscriber line) technology
has enabled residential users to access the Internet at tens of Mbps over twisted pair
(when users live close to the ISP’s modem).
Coaxial Cable
Like twisted pair, coaxial cable consists of two copper conductors, but the two conductors are concentric rather than parallel. With this construction and special insulation and shielding, coaxial cable can achieve high data transmission rates. Coaxial
cable is quite common in cable television systems. As we saw earlier, cable television systems have recently been coupled with cable modems to provide residential
users with Internet access at rates of tens of Mbps. In cable television and cable
Internet access, the transmitter shifts the digital signal to a specific frequency band,
and the resulting analog signal is sent from the transmitter to one or more receivers.
Coaxial cable can be used as a guided shared medium. Specifically, a number of
end systems can be connected directly to the cable, with each of the end systems
receiving whatever is sent by the other end systems.
Fiber Optics
An optical fiber is a thin, flexible medium that conducts pulses of light, with each
pulse representing a bit. A single optical fiber can support tremendous bit rates, up
to tens or even hundreds of gigabits per second. They are immune to electromagnetic interference, have very low signal attenuation up to 100 kilometers, and are
very hard to tap. These characteristics have made fiber optics the preferred longhaul guided transmission media, particularly for overseas links. Many of the longdistance telephone networks in the United States and elsewhere now use fiber optics
exclusively. Fiber optics is also prevalent in the backbone of the Internet. However,
the high cost of optical devices—such as transmitters, receivers, and switches—has
hindered their deployment for short-haul transport, such as in a LAN or into the
1.2
•
THE NETWORK EDGE
home in a residential access network. The Optical Carrier (OC) standard link speeds
range from 51.8 Mbps to 39.8 Gbps; these specifications are often referred to as OCn, where the link speed equals n × 51.8 Mbps. Standards in use today include OC-1,
OC-3, OC-12, OC-24, OC-48, OC-96, OC-192, OC-768. [Mukherjee 2006,
Ramaswamy 2010] provide coverage of various aspects of optical networking.
Terrestrial Radio Channels
Radio channels carry signals in the electromagnetic spectrum. They are an attractive
medium because they require no physical wire to be installed, can penetrate walls,
provide connectivity to a mobile user, and can potentially carry a signal for long distances. The characteristics of a radio channel depend significantly on the propagation
environment and the distance over which a signal is to be carried. Environmental considerations determine path loss and shadow fading (which decrease the signal strength
as the signal travels over a distance and around/through obstructing objects), multipath fading (due to signal reflection off of interfering objects), and interference (due
to other transmissions and electromagnetic signals).
Terrestrial radio channels can be broadly classified into three groups: those that
operate over very short distance (e.g., with one or two meters); those that operate in
local areas, typically spanning from ten to a few hundred meters; and those that
operate in the wide area, spanning tens of kilometers. Personal devices such as wireless headsets, keyboards, and medical devices operate over short distances; the
wireless LAN technologies described in Section 1.2.1 use local-area radio channels;
the cellular access technologies use wide-area radio channels. We’ll discuss radio
channels in detail in Chapter 6.
Satellite Radio Channels
A communication satellite links two or more Earth-based microwave transmitter/
receivers, known as ground stations. The satellite receives transmissions on one frequency band, regenerates the signal using a repeater (discussed below), and transmits
the signal on another frequency. Two types of satellites are used in communications:
geostationary satellites and low-earth orbiting (LEO) satellites.
Geostationary satellites permanently remain above the same spot on Earth. This
stationary presence is achieved by placing the satellite in orbit at 36,000 kilometers
above Earth’s surface. This huge distance from ground station through satellite back
to ground station introduces a substantial signal propagation delay of 280 milliseconds. Nevertheless, satellite links, which can operate at speeds of hundreds of Mbps,
are often used in areas without access to DSL or cable-based Internet access.
LEO satellites are placed much closer to Earth and do not remain permanently
above one spot on Earth. They rotate around Earth (just as the Moon does) and may
communicate with each other, as well as with ground stations. To provide continuous
21
22
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
coverage to an area, many satellites need to be placed in orbit. There are currently
many low-altitude communication systems in development. Lloyd’s satellite constellations Web page [Wood 2012] provides and collects information on satellite
constellation systems for communications. LEO satellite technology may be used
for Internet access sometime in the future.
1.3 The Network Core
Having examined the Internet’s edge, let us now delve more deeply inside the network core—the mesh of packet switches and links that interconnects the Internet’s
end systems. Figure 1.10 highlights the network core with thick, shaded lines.
1.3.1 Packet Switching
In a network application, end systems exchange messages with each other. Messages can contain anything the application designer wants. Messages may perform a
control function (for example, the “Hi” messages in our handshaking example in
Figure 1.2) or can contain data, such as an email message, a JPEG image, or an MP3
audio file. To send a message from a source end system to a destination end system,
the source breaks long messages into smaller chunks of data known as packets.
Between source and destination, each packet travels through communication links
and packet switches (for which there are two predominant types, routers and linklayer switches). Packets are transmitted over each communication link at a rate
equal to the full transmission rate of the link. So, if a source end system or a packet
switch is sending a packet of L bits over a link with transmission rate R bits/sec, then
the time to transmit the packet is L/R seconds.
Store-and-Forward Transmission
Most packet switches use store-and-forward transmission at the inputs to the
links. Store-and-forward transmission means that the packet switch must receive
the entire packet before it can begin to transmit the first bit of the packet onto the
outbound link. To explore store-and-forward transmission in more detail, consider
a simple network consisting of two end systems connected by a single router, as
shown in Figure 1.11. A router will typically have many incident links, since its job
is to switch an incoming packet onto an outgoing link; in this simple example, the
router has the rather simple task of transferring a packet from one (input) link to
the only other attached link. In this example, the source has three packets, each
consisting of L bits, to send to the destination. At the snapshot of time shown in
Figure 1.11, the source has transmitted some of packet 1, and the front of packet 1
has already arrived at the router. Because the router employs store-and-forwarding,
at this instant of time, the router cannot transmit the bits it has received; instead it
1.3
National or
Global ISP
Mobile Network
Local or
Regional ISP
Home Network
Enterprise Network
Figure 1.10
The network core
•
THE NETWORK CORE
23
24
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
321
R bps
Front of packet 1
stored in router,
awaiting remaining
bits before forwarding
Source
Figure 1.11
Destination
Store-and-forward packet switching
must first buffer (i.e., “store”) the packet’s bits. Only after the router has received
all of the packet’s bits can it begin to transmit (i.e., “forward”) the packet onto the
outbound link. To gain some insight into store-and-forward transmission, let’s now
calculate the amount of time that elapses from when the source begins to send the
packet until the destination has received the entire packet. (Here we will ignore
propagation delay—the time it takes for the bits to travel across the wire at near the
speed of light—which will be discussed in Section 1.4.) The source begins to transmit at time 0; at time L/R seconds, the source has transmitted the entire packet, and
the entire packet has been received and stored at the router (since there is no propagation delay). At time L/R seconds, since the router has just received the entire
packet, it can begin to transmit the packet onto the outbound link towards the destination; at time 2L/R, the router has transmitted the entire packet, and the entire
packet has been received by the destination. Thus, the total delay is 2L/R. If
the switch instead forwarded bits as soon as they arrive (without first receiving the
entire packet), then the total delay would be L/R since bits are not held up at
the router. But, as we will discuss in Section 1.4, routers need to receive, store, and
process the entire packet before forwarding.
Now let’s calculate the amount of time that elapses from when the source
begins to send the first packet until the destination has received all three packets.
As before, at time L/R, the router begins to forward the first packet. But also at time
L/R the source will begin to send the second packet, since it has just finished sending the entire first packet. Thus, at time 2L/R, the destination has received the first
packet and the router has received the second packet. Similarly, at time 3L/R, the
destination has received the first two packets and the router has received the third
packet. Finally, at time 4L/R the destination has received all three packets!
Let’s now consider the general case of sending one packet from source to destination over a path consisting of N links each of rate R (thus, there are N-1 routers
between source and destination). Applying the same logic as above, we see that the
end-to-end delay is:
[email protected]@end = N
L
R
(1.1)
You may now want to try to determine what the delay would be for P packets sent
over a series of N links.
1.3
•
THE NETWORK CORE
Queuing Delays and Packet Loss
Each packet switch has multiple links attached to it. For each attached link, the
packet switch has an output buffer (also called an output queue), which stores
packets that the router is about to send into that link. The output buffers play a key
role in packet switching. If an arriving packet needs to be transmitted onto a link but
finds the link busy with the transmission of another packet, the arriving packet must
wait in the output buffer. Thus, in addition to the store-and-forward delays, packets
suffer output buffer queuing delays. These delays are variable and depend on the
level of congestion in the network. Since the amount of buffer space is finite, an
arriving packet may find that the buffer is completely full with other packets waiting for transmission. In this case, packet loss will occur—either the arriving packet
or one of the already-queued packets will be dropped.
Figure 1.12 illustrates a simple packet-switched network. As in Figure 1.11,
packets are represented by three-dimensional slabs. The width of a slab represents
the number of bits in the packet. In this figure, all packets have the same width and
hence the same length. Suppose Hosts A and B are sending packets to Host E. Hosts
A and B first send their packets along 10 Mbps Ethernet links to the first router. The
router then directs these packets to the 1.5 Mbps link. If, during a short interval of
time, the arrival rate of packets to the router (when converted to bits per second)
exceeds 1.5 Mbps, congestion will occur at the router as packets queue in the link’s
output buffer before being transmitted onto the link. For example, if Host A and B
each send a burst of five packets back-to-back at the same time, then most of these
packets will spend some time waiting in the queue. The situation is, in fact, entirely
analogous to many common-day situations—for example, when we wait in line for
a bank teller or wait in front of a tollbooth. We’ll examine this queuing delay in
more detail in Section 1.4.
10 Mbps Ethernet
C
A
1.5 Mbps
Queue of
packets waiting
for output link
B
Key:
D
Packets
Figure 1.12
Packet switching
E
25
26
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
Forwarding Tables and Routing Protocols
Earlier, we said that a router takes a packet arriving on one of its attached
communication links and forwards that packet onto another one of its attached communication links. But how does the router determine which link it should forward
the packet onto? Packet forwarding is actually done in different ways in different
types of computer networks. Here, we briefly describe how it is done in the
Internet.
In the Internet, every end system has an address called an IP address. When a
source end system wants to send a packet to a destination end system, the source
includes the destination’s IP address in the packet’s header. As with postal addresses,
this address has a hierarchical structure. When a packet arrives at a router in the
network, the router examines a portion of the packet’s destination address and forwards the packet to an adjacent router. More specifically, each router has a
forwarding table that maps destination addresses (or portions of the destination
addresses) to that router’s outbound links. When a packet arrives at a router, the
router examines the address and searches its forwarding table, using this destination
address, to find the appropriate outbound link. The router then directs the packet to
this outbound link.
The end-to-end routing process is analogous to a car driver who does not use
maps but instead prefers to ask for directions. For example, suppose Joe is driving
from Philadelphia to 156 Lakeside Drive in Orlando, Florida. Joe first drives to his
neighborhood gas station and asks how to get to 156 Lakeside Drive in Orlando,
Florida. The gas station attendant extracts the Florida portion of the address and
tells Joe that he needs to get onto the interstate highway I-95 South, which has an
entrance just next to the gas station. He also tells Joe that once he enters Florida,
he should ask someone else there. Joe then takes I-95 South until he gets to Jacksonville, Florida, at which point he asks another gas station attendant for directions.
The attendant extracts the Orlando portion of the address and tells Joe that he
should continue on I-95 to Daytona Beach and then ask someone else. In Daytona
Beach, another gas station attendant also extracts the Orlando portion of the
address and tells Joe that he should take I-4 directly to Orlando. Joe takes I-4 and
gets off at the Orlando exit. Joe goes to another gas station attendant, and this time
the attendant extracts the Lakeside Drive portion of the address and tells Joe the
road he must follow to get to Lakeside Drive. Once Joe reaches Lakeside Drive, he
asks a kid on a bicycle how to get to his destination. The kid extracts the 156 portion of the address and points to the house. Joe finally reaches his ultimate destination. In the above analogy, the gas station attendants and kids on bicycles are
analogous to routers.
We just learned that a router uses a packet’s destination address to index a forwarding table and determine the appropriate outbound link. But this statement begs
yet another question: How do forwarding tables get set? Are they configured by
1.3
•
THE NETWORK CORE
hand in each and every router, or does the Internet use a more automated procedure?
This issue will be studied in depth in Chapter 4. But to whet your appetite here,
we’ll note now that the Internet has a number of special routing protocols that are
used to automatically set the forwarding tables. A routing protocol may, for example, determine the shortest path from each router to each destination and use the
shortest path results to configure the forwarding tables in the routers.
How would you actually like to see the end-to-end route that packets take in the
Internet? We now invite you to get your hands dirty by interacting with the Traceroute program. Simply visit the site www.traceroute.org, choose a source in a particular country, and trace the route from that source to your computer. (For a discussion of
Traceroute, see Section 1.4.)
1.3.2 Circuit Switching
There are two fundamental approaches to moving data through a network of links
and switches: circuit switching and packet switching. Having covered packetswitched networks in the previous subsection, we now turn our attention to circuitswitched networks.
In circuit-switched networks, the resources needed along a path (buffers, link
transmission rate) to provide for communication between the end systems are
reserved for the duration of the communication session between the end systems. In
packet-switched networks, these resources are not reserved; a session’s messages
use the resources on demand, and as a consequence, may have to wait (that is,
queue) for access to a communication link. As a simple analogy, consider two
restaurants, one that requires reservations and another that neither requires reservations nor accepts them. For the restaurant that requires reservations, we have to go
through the hassle of calling before we leave home. But when we arrive at the
restaurant we can, in principle, immediately be seated and order our meal. For the
restaurant that does not require reservations, we don’t need to bother to reserve a
table. But when we arrive at the restaurant, we may have to wait for a table before
we can be seated.
Traditional telephone networks are examples of circuit-switched networks.
Consider what happens when one person wants to send information (voice or facsimile) to another over a telephone network. Before the sender can send the information, the network must establish a connection between the sender and the
receiver. This is a bona fide connection for which the switches on the path
between the sender and receiver maintain connection state for that connection. In
the jargon of telephony, this connection is called a circuit. When the network
establishes the circuit, it also reserves a constant transmission rate in the network’s links (representing a fraction of each link’s transmission capacity) for the
duration of the connection. Since a given transmission rate has been reserved for
this sender-to-receiver connection, the sender can transfer the data to the receiver
at the guaranteed constant rate.
27
28
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
Figure 1.13 illustrates a circuit-switched network. In this network, the four circuit switches are interconnected by four links. Each of these links has four circuits,
so that each link can support four simultaneous connections. The hosts (for example, PCs and workstations) are each directly connected to one of the switches. When
two hosts want to communicate, the network establishes a dedicated end-to-end
connection between the two hosts. Thus, in order for Host A to communicate with
Host B, the network must first reserve one circuit on each of two links. In this example, the dedicated end-to-end connection uses the second circuit in the first link and
the fourth circuit in the second link. Because each link has four circuits, for each
link used by the end-to-end connection, the connection gets one fourth of the link’s
total transmission capacity for the duration of the connection. Thus, for example, if
each link between adjacent switches has a transmission rate of 1 Mbps, then each
end-to-end circuit-switch connection gets 250 kbps of dedicated transmission rate.
In contrast, consider what happens when one host wants to send a packet to
another host over a packet-switched network, such as the Internet. As with circuit
switching, the packet is transmitted over a series of communication links. But different
from circuit switching, the packet is sent into the network without reserving any link
resources whatsoever. If one of the links is congested because other packets need to be
transmitted over the link at the same time, then the packet will have to wait in a buffer
at the sending side of the transmission link and suffer a delay. The Internet makes its
best effort to deliver packets in a timely manner, but it does not make any guarantees.
Multiplexing in Circuit-Switched Networks
A circuit in a link is implemented with either frequency-division multiplexing
(FDM) or time-division multiplexing (TDM). With FDM, the frequency spectrum of a link is divided up among the connections established across the link.
Figure 1.13
A simple circuit-switched network consisting of
four switches and four links
1.3
•
THE NETWORK CORE
Specifically, the link dedicates a frequency band to each connection for the
duration of the connection. In telephone networks, this frequency band typically
has a width of 4 kHz (that is, 4,000 hertz or 4,000 cycles per second). The width
of the band is called, not surprisingly, the bandwidth. FM radio stations also use
FDM to share the frequency spectrum between 88 MHz and 108 MHz, with each
station being allocated a specific frequency band.
For a TDM link, time is divided into frames of fixed duration, and each frame
is divided into a fixed number of time slots. When the network establishes a connection across a link, the network dedicates one time slot in every frame to this connection. These slots are dedicated for the sole use of that connection, with one time slot
available for use (in every frame) to transmit the connection’s data.
Figure 1.14 illustrates FDM and TDM for a specific network link supporting up
to four circuits. For FDM, the frequency domain is segmented into four bands, each
of bandwidth 4 kHz. For TDM, the time domain is segmented into frames, with four
time slots in each frame; each circuit is assigned the same dedicated slot in the
revolving TDM frames. For TDM, the transmission rate of a circuit is equal to the
frame rate multiplied by the number of bits in a slot. For example, if the link transmits 8,000 frames per second and each slot consists of 8 bits, then the transmission
rate of a circuit is 64 kbps.
Proponents of packet switching have always argued that circuit switching is
wasteful because the dedicated circuits are idle during silent periods. For example,
FDM
4KHz
Link
Frequency
4KHz
TDM
1 2
Slot
3
4 1 2
3
4
1 2
3
4 1 2
3
4
Frame
Time
Key:
2
Figure 1.14
All slots labeled “2” are dedicated
to a specific sender-receiver pair.
With FDM, each circuit continuously gets a fraction of the
bandwidth. With TDM, each circuit gets all of the bandwidth
periodically during brief intervals of time (that is, during slots)
29
30
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
when one person in a telephone call stops talking, the idle network resources (frequency bands or time slots in the links along the connection’s route) cannot be used
by other ongoing connections. As another example of how these resources can be
underutilized, consider a radiologist who uses a circuit-switched network to
remotely access a series of x-rays. The radiologist sets up a connection, requests an
image, contemplates the image, and then requests a new image. Network resources
are allocated to the connection but are not used (i.e., are wasted) during the radiologist’s contemplation periods. Proponents of packet switching also enjoy pointing out
that establishing end-to-end circuits and reserving end-to-end transmission capacity
is complicated and requires complex signaling software to coordinate the operation
of the switches along the end-to-end path.
Before we finish our discussion of circuit switching, let’s work through a numerical example that should shed further insight on the topic. Let us consider how long it
takes to send a file of 640,000 bits from Host A to Host B over a circuit-switched network. Suppose that all links in the network use TDM with 24 slots and have a bit rate
of 1.536 Mbps. Also suppose that it takes 500 msec to establish an end-to-end circuit
before Host A can begin to transmit the file. How long does it take to send the file?
Each circuit has a transmission rate of (1.536 Mbps)/24 = 64 kbps, so it takes (640,000
bits)/(64 kbps) = 10 seconds to transmit the file. To this 10 seconds we add the circuit
establishment time, giving 10.5 seconds to send the file. Note that the transmission
time is independent of the number of links: The transmission time would be 10 seconds if the end-to-end circuit passed through one link or a hundred links. (The actual
end-to-end delay also includes a propagation delay; see Section 1.4.)
Packet Switching Versus Circuit Switching
Having described circuit switching and packet switching, let us compare the two.
Critics of packet switching have often argued that packet switching is not suitable
for real-time services (for example, telephone calls and video conference calls)
because of its variable and unpredictable end-to-end delays (due primarily to variable and unpredictable queuing delays). Proponents of packet switching argue that
(1) it offers better sharing of transmission capacity than circuit switching and (2) it
is simpler, more efficient, and less costly to implement than circuit switching. An
interesting discussion of packet switching versus circuit switching is [MolineroFernandez 2002]. Generally speaking, people who do not like to hassle with restaurant reservations prefer packet switching to circuit switching.
Why is packet switching more efficient? Let’s look at a simple example. Suppose users share a 1 Mbps link. Also suppose that each user alternates between periods of activity, when a user generates data at a constant rate of 100 kbps, and periods
of inactivity, when a user generates no data. Suppose further that a user is active
only 10 percent of the time (and is idly drinking coffee during the remaining 90 percent of the time). With circuit switching, 100 kbps must be reserved for each user at
1.3
•
THE NETWORK CORE
all times. For example, with circuit-switched TDM, if a one-second frame is divided
into 10 time slots of 100 ms each, then each user would be allocated one time slot
per frame.
Thus, the circuit-switched link can support only 10 (= 1 Mbps/100 kbps) simultaneous users. With packet switching, the probability that a specific user is active is
0.1 (that is, 10 percent). If there are 35 users, the probability that there are 11 or
more simultaneously active users is approximately 0.0004. (Homework Problem P8
outlines how this probability is obtained.) When there are 10 or fewer simultaneously active users (which happens with probability 0.9996), the aggregate arrival
rate of data is less than or equal to 1 Mbps, the output rate of the link. Thus, when
there are 10 or fewer active users, users’ packets flow through the link essentially
without delay, as is the case with circuit switching. When there are more than 10
simultaneously active users, then the aggregate arrival rate of packets exceeds the
output capacity of the link, and the output queue will begin to grow. (It continues to
grow until the aggregate input rate falls back below 1 Mbps, at which point the
queue will begin to diminish in length.) Because the probability of having more than
10 simultaneously active users is minuscule in this example, packet switching provides essentially the same performance as circuit switching, but does so while
allowing for more than three times the number of users.
Let’s now consider a second simple example. Suppose there are 10 users and that
one user suddenly generates one thousand 1,000-bit packets, while other users
remain quiescent and do not generate packets. Under TDM circuit switching with 10
slots per frame and each slot consisting of 1,000 bits, the active user can only use its
one time slot per frame to transmit data, while the remaining nine time slots in each
frame remain idle. It will be 10 seconds before all of the active user’s one million bits
of data has been transmitted. In the case of packet switching, the active user can continuously send its packets at the full link rate of 1 Mbps, since there are no other users
generating packets that need to be multiplexed with the active user’s packets. In this
case, all of the active user’s data will be transmitted within 1 second.
The above examples illustrate two ways in which the performance of packet
switching can be superior to that of circuit switching. They also highlight the crucial
difference between the two forms of sharing a link’s transmission rate among multiple data streams. Circuit switching pre-allocates use of the transmission link regardless of demand, with allocated but unneeded link time going unused. Packet
switching on the other hand allocates link use on demand. Link transmission capacity will be shared on a packet-by-packet basis only among those users who have
packets that need to be transmitted over the link.
Although packet switching and circuit switching are both prevalent in today’s
telecommunication networks, the trend has certainly been in the direction of packet
switching. Even many of today’s circuit-switched telephone networks are slowly
migrating toward packet switching. In particular, telephone networks often use
packet switching for the expensive overseas portion of a telephone call.
31
32
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
1.3.3 A Network of Networks
We saw earlier that end systems (PCs, smartphones, Web servers, mail servers,
and so on) connect into the Internet via an access ISP. The access ISP can provide either wired or wireless connectivity, using an array of access technologies
including DSL, cable, FTTH, Wi-Fi, and cellular. Note that the access ISP does
not have to be a telco or a cable company; instead it can be, for example, a university (providing Internet access to students, staff, and faculty), or a company
(providing access for its employees). But connecting end users and content
providers into an access ISP is only a small piece of solving the puzzle of connecting the billions of end systems that make up the Internet. To complete this
puzzle, the access ISPs themselves must be interconnected. This is done by creating a network of networks—understanding this phrase is the key to understanding the Internet.
Over the years, the network of networks that forms the Internet has evolved into
a very complex structure. Much of this evolution is driven by economics and
national policy, rather than by performance considerations. In order to understand
today’s Internet network structure, let’s incrementally build a series of network
structures, with each new structure being a better approximation of the complex
Internet that we have today. Recall that the overarching goal is to interconnect the
access ISPs so that all end systems can send packets to each other. One naive
approach would be to have each access ISP directly connect with every other access
ISP. Such a mesh design is, of course, much too costly for the access ISPs, as it
would require each access ISP to have a separate communication link to each of the
hundreds of thousands of other access ISPs all over the world.
Our first network structure, Network Structure 1, interconnects all of the
access ISPs with a single global transit ISP. Our (imaginary) global transit ISP is
a network of routers and communication links that not only spans the globe, but
also has at least one router near each of the hundreds of thousands of access ISPs.
Of course, it would be very costly for the global ISP to build such an extensive
network. To be profitable, it would naturally charge each of the access ISPs for
connectivity, with the pricing reflecting (but not necessarily directly proportional
to) the amount of traffic an access ISP exchanges with the global ISP. Since the
access ISP pays the global transit ISP, the access ISP is said to be a customer and
the global transit ISP is said to be a provider.
Now if some company builds and operates a global transit ISP that is profitable,
then it is natural for other companies to build their own global transit ISPs and compete with the original global transit ISP. This leads to Network Structure 2, which
consists of the hundreds of thousands of access ISPs and multiple global transit
ISPs. The access ISPs certainly prefer Network Structure 2 over Network Structure
1 since they can now choose among the competing global transit providers as a
function of their pricing and services. Note, however, that the global transit ISPs
1.3
•
THE NETWORK CORE
themselves must interconnect: Otherwise access ISPs connected to one of the global
transit providers would not be able to communicate with access ISPs connected to
the other global transit providers.
Network Structure 2, just described, is a two-tier hierarchy with global transit
providers residing at the top tier and access ISPs at the bottom tier. This assumes
that global transit ISPs are not only capable of getting close to each and every access
ISP, but also find it economically desirable to do so. In reality, although some ISPs
do have impressive global coverage and do directly connect with many access ISPs,
no ISP has presence in each and every city in the world. Instead, in any given region,
there may be a regional ISP to which the access ISPs in the region connect. Each
regional ISP then connects to tier-1 ISPs. Tier-1 ISPs are similar to our (imaginary)
global transit ISP; but tier-1 ISPs, which actually do exist, do not have a presence in
every city in the world. There are approximately a dozen tier-1 ISPs, including
Level 3 Communications, AT&T, Sprint, and NTT. Interestingly, no group officially
sanctions tier-1 status; as the saying goes—if you have to ask if you’re a member of
a group, you’re probably not.
Returning to this network of networks, not only are there multiple competing tier1 ISPs, there may be multiple competing regional ISPs in a region. In such a hierarchy, each access ISP pays the regional ISP to which it connects, and each regional ISP
pays the tier-1 ISP to which it connects. (An access ISP can also connect directly to a
tier-1 ISP, in which case it pays the tier-1 ISP). Thus, there is customer-provider
relationship at each level of the hierarchy. Note that the tier-1 ISPs do not pay anyone
as they are at the top of the hierarchy. To further complicate matters, in some regions,
there may be a larger regional ISP (possibly spanning an entire country) to which the
smaller regional ISPs in that region connect; the larger regional ISP then connects to a
tier-1 ISP. For example, in China, there are access ISPs in each city, which connect to
provincial ISPs, which in turn connect to national ISPs, which finally connect to tier-1
ISPs [Tian 2012]. We refer to this multi-tier hierarchy, which is still only a crude
approximation of today’s Internet, as Network Structure 3.
To build a network that more closely resembles today’s Internet, we must add
points of presence (PoPs), multi-homing, peering, and Internet exchange points
(IXPs) to the hierarchical Network Structure 3. PoPs exist in all levels of the hierarchy, except for the bottom (access ISP) level. A PoP is simply a group of one or
more routers (at the same location) in the provider’s network where customer ISPs
can connect into the provider ISP. For a customer network to connect to a
provider’s PoP, it can lease a high-speed link from a third-party telecommunications provider to directly connect one of its routers to a router at the PoP. Any ISP
(except for tier-1 ISPs) may choose to multi-home, that is, to connect to two or
more provider ISPs. So, for example, an access ISP may multi-home with two
regional ISPs, or it may multi-home with two regional ISPs and also with a tier-1
ISP. Similarly, a regional ISP may multi-home with multiple tier-1 ISPs. When an
33
34
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
ISP multi-homes, it can continue to send and receive packets into the Internet even
if one of its providers has a failure.
As we just learned, customer ISPs pay their provider ISPs to obtain global
Internet interconnectivity. The amount that a customer ISP pays a provider ISP
reflects the amount of traffic it exchanges with the provider. To reduce these costs, a
pair of nearby ISPs at the same level of the hierarchy can peer, that is, they can
directly connect their networks together so that all the traffic between them passes
over the direct connection rather than through upstream intermediaries. When two
ISPs peer, it is typically settlement-free, that is, neither ISP pays the other. As noted
earlier, tier-1 ISPs also peer with one another, settlement-free. For a readable discussion of peering and customer-provider relationships, see [Van der Berg 2008].
Along these same lines, a third-party company can create an Internet Exchange
Point (IXP) (typically in a stand-alone building with its own switches), which is a
meeting point where multiple ISPs can peer together. There are roughly 300 IXPs in
the Internet today [Augustin 2009]. We refer to this ecosystem—consisting of
access ISPs, regional ISPs, tier-1 ISPs, PoPs, multi-homing, peering, and IXPs—as
Network Structure 4.
We now finally arrive at Network Structure 5, which describes the Internet of
2012. Network Structure 5, illustrated in Figure 1.15, builds on top of Network
Structure 4 by adding content provider networks. Google is currently one of the
leading examples of such a content provider network. As of this writing, it is estimated that Google has 30 to 50 data centers distributed across North America,
Europe, Asia, South America, and Australia. Some of these data centers house
over one hundred thousand servers, while other data centers are smaller, housing
only hundreds of servers. The Google data centers are all interconnected via
Google’s private TCP/IP network, which spans the entire globe but is nevertheless
separate from the public Internet. Importantly, the Google private network only
Tier 1
ISP
Content provider
(e.g., Google)
Tier 1
ISP
IXP
IXP
Regional
ISP
access
ISP
access
ISP
Figure 1.15
access
ISP
IXP
Regional
ISP
access
ISP
Interconnection of ISPs
access
ISP
access
ISP
access
ISP
access
ISP
1.4
•
DELAY, LOSS, AND THROUGHPUT IN PACKET-SWITCHED NETWORKS
carries traffic to/from Google servers. As shown in Figure 1.15, the Google
private network attempts to “bypass” the upper tiers of the Internet by peering
(settlement free) with lower-tier ISPs, either by directly connecting with them or
by connecting with them at IXPs [Labovitz 2010]. However, because many access
ISPs can still only be reached by transiting through tier-1 networks, the Google
network also connects to tier-1 ISPs, and pays those ISPs for the traffic it
exchanges with them. By creating its own network, a content provider not only
reduces its payments to upper-tier ISPs, but also has greater control of how its
services are ultimately delivered to end users. Google’s network infrastructure is
described in greater detail in Section 7.2.4.
In summary, today’s Internet—a network of networks—is complex, consisting of
a dozen or so tier-1 ISPs and hundreds of thousands of lower-tier ISPs. The ISPs are
diverse in their coverage, with some spanning multiple continents and oceans, and
others limited to narrow geographic regions. The lower-tier ISPs connect to the
higher-tier ISPs, and the higher-tier ISPs interconnect with one another. Users and
content providers are customers of lower-tier ISPs, and lower-tier ISPs are customers
of higher-tier ISPs. In recent years, major content providers have also created their
own networks and connect directly into lower-tier ISPs where possible.
1.4 Delay, Loss, and Throughput
in Packet-Switched Networks
Back in Section 1.1 we said that the Internet can be viewed as an infrastructure
that provides services to distributed applications running on end systems. Ideally,
we would like Internet services to be able to move as much data as we want
between any two end systems, instantaneously, without any loss of data. Alas, this
is a lofty goal, one that is unachievable in reality. Instead, computer networks necessarily constrain throughput (the amount of data per second that can be transferred) between end systems, introduce delays between end systems, and can
actually lose packets. On one hand, it is unfortunate that the physical laws of reality introduce delay and loss as well as constrain throughput. On the other hand,
because computer networks have these problems, there are many fascinating
issues surrounding how to deal with the problems—more than enough issues to
fill a course on computer networking and to motivate thousands of PhD theses! In
this section, we’ll begin to examine and quantify delay, loss, and throughput in
computer networks.
1.4.1 Overview of Delay in Packet-Switched Networks
Recall that a packet starts in a host (the source), passes through a series of routers,
and ends its journey in another host (the destination). As a packet travels from one
node (host or router) to the subsequent node (host or router) along this path, the
35
36
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
A
B
Propagation
Nodal
Queueing Transmission
processing (waiting for
transmission)
Figure 1.16
The nodal delay at router A
packet suffers from several types of delays at each node along the path. The most
important of these delays are the nodal processing delay, queuing delay, transmission delay, and propagation delay; together, these delays accumulate to give a total
nodal delay. The performance of many Internet applications—such as search, Web
browsing, email, maps, instant messaging, and voice-over-IP—are greatly affected
by network delays. In order to acquire a deep understanding of packet switching and
computer networks, we must understand the nature and importance of these delays.
Types of Delay
Let’s explore these delays in the context of Figure 1.16. As part of its end-to-end
route between source and destination, a packet is sent from the upstream node
through router A to router B. Our goal is to characterize the nodal delay at router A.
Note that router A has an outbound link leading to router B. This link is preceded by
a queue (also known as a buffer). When the packet arrives at router A from the
upstream node, router A examines the packet’s header to determine the appropriate
outbound link for the packet and then directs the packet to this link. In this example,
the outbound link for the packet is the one that leads to router B. A packet can be
transmitted on a link only if there is no other packet currently being transmitted on
the link and if there are no other packets preceding it in the queue; if the link is
currently busy or if there are other packets already queued for the link, the newly
arriving packet will then join the queue.
Processing Delay
The time required to examine the packet’s header and determine where to direct the
packet is part of the processing delay. The processing delay can also include other
factors, such as the time needed to check for bit-level errors in the packet that occurred
in transmitting the packet’s bits from the upstream node to router A. Processing delays
1.4
•
DELAY, LOSS, AND THROUGHPUT IN PACKET-SWITCHED NETWORKS
in high-speed routers are typically on the order of microseconds or less. After this
nodal processing, the router directs the packet to the queue that precedes the link to
router B. (In Chapter 4 we’ll study the details of how a router operates.)
Queuing Delay
At the queue, the packet experiences a queuing delay as it waits to be transmitted onto
the link. The length of the queuing delay of a specific packet will depend on the number of earlier-arriving packets that are queued and waiting for transmission onto the
link. If the queue is empty and no other packet is currently being transmitted, then our
packet’s queuing delay will be zero. On the other hand, if the traffic is heavy and many
other packets are also waiting to be transmitted, the queuing delay will be long. We
will see shortly that the number of packets that an arriving packet might expect to find
is a function of the intensity and nature of the traffic arriving at the queue. Queuing
delays can be on the order of microseconds to milliseconds in practice.
Transmission Delay
Assuming that packets are transmitted in a first-come-first-served manner, as is common in packet-switched networks, our packet can be transmitted only after all the
packets that have arrived before it have been transmitted. Denote the length of the
packet by L bits, and denote the transmission rate of the link from router A to router
B by R bits/sec. For example, for a 10 Mbps Ethernet link, the rate is R = 10 Mbps;
for a 100 Mbps Ethernet link, the rate is R = 100 Mbps. The transmission delay is
L/R. This is the amount of time required to push (that is, transmit) all of the packet’s
bits into the link. Transmission delays are typically on the order of microseconds to
milliseconds in practice.
Propagation Delay
Once a bit is pushed into the link, it needs to propagate to router B. The time
required to propagate from the beginning of the link to router B is the propagation
delay. The bit propagates at the propagation speed of the link. The propagation
speed depends on the physical medium of the link (that is, fiber optics, twisted-pair
copper wire, and so on) and is in the range of
2 ⭈108 meters/sec to 3 ⭈108 meters/sec
which is equal to, or a little less than, the speed of light. The propagation delay is
the distance between two routers divided by the propagation speed. That is, the
propagation delay is d/s, where d is the distance between router A and router B and s
is the propagation speed of the link. Once the last bit of the packet propagates to
node B, it and all the preceding bits of the packet are stored in router B. The whole
37
38
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
process then continues with router B now performing the forwarding. In wide-area
networks, propagation delays are on the order of milliseconds.
Comparing Transmission and Propagation Delay
Newcomers to the field of computer networking sometimes have difficulty understanding the difference between transmission delay and propagation delay. The difference is subtle but important. The transmission delay is the amount of time required for
the router to push out the packet; it is a function of the packet’s length and the transmission rate of the link, but has nothing to do with the distance between the two
routers. The propagation delay, on the other hand, is the time it takes a bit to propagate
from one router to the next; it is a function of the distance between the two routers, but
has nothing to do with the packet’s length or the transmission rate of the link.
An analogy might clarify the notions of transmission and propagation delay. Consider a highway that has a tollbooth every 100 kilometers, as shown in Figure 1.17.
You can think of the highway segments between tollbooths as links and the tollbooths as routers. Suppose that cars travel (that is, propagate) on the highway at a
rate of 100 km/hour (that is, when a car leaves a tollbooth, it instantaneously accelerates to 100 km/hour and maintains that speed between tollbooths). Suppose next
that 10 cars, traveling together as a caravan, follow each other in a fixed order. You
can think of each car as a bit and the caravan as a packet. Also suppose that each
tollbooth services (that is, transmits) a car at a rate of one car per 12 seconds, and
that it is late at night so that the caravan’s cars are the only cars on the highway.
Finally, suppose that whenever the first car of the caravan arrives at a tollbooth, it
waits at the entrance until the other nine cars have arrived and lined up behind it.
(Thus the entire caravan must be stored at the tollbooth before it can begin to be forwarded.) The time required for the tollbooth to push the entire caravan onto the
highway is (10 cars)/(5 cars/minute) = 2 minutes. This time is analogous to the
transmission delay in a router. The time required for a car to travel from the exit of
one tollbooth to the next tollbooth is 100 km/(100 km/hour) = 1 hour. This time is
analogous to propagation delay. Therefore, the time from when the caravan is stored
in front of a tollbooth until the caravan is stored in front of the next tollbooth is the
sum of transmission delay and propagation delay—in this example, 62 minutes.
100 km
Ten-car
caravan
Figure 1.17
Caravan analogy
Toll
booth
100 km
Toll
booth
1.4
•
DELAY, LOSS, AND THROUGHPUT IN PACKET-SWITCHED NETWORKS
Let’s explore this analogy a bit more. What would happen if the tollbooth service
time for a caravan were greater than the time for a car to travel between tollbooths?
For example, suppose now that the cars travel at the rate of 1,000 km/hour and the tollbooth services cars at the rate of one car per minute. Then the traveling delay between
two tollbooths is 6 minutes and the time to serve a caravan is 10 minutes. In this case,
the first few cars in the caravan will arrive at the second tollbooth before the last cars
in the caravan leave the first tollbooth. This situation also arises in packet-switched
networks—the first bits in a packet can arrive at a router while many of the remaining
bits in the packet are still waiting to be transmitted by the preceding router.
If a picture speaks a thousand words, then an animation must speak a million
words. The companion Web site for this textbook provides an interactive Java applet
that nicely illustrates and contrasts transmission delay and propagation delay. The
reader is highly encouraged to visit that applet. [Smith 2009] also provides a very
readable discussion of propagation, queueing, and transmission delays.
If we let dproc, dqueue, dtrans, and dprop denote the processing, queuing, transmission, and propagation delays, then the total nodal delay is given by
dnodal = dproc + dqueue + dtrans + dprop
The contribution of these delay components can vary significantly. For example,
dprop can be negligible (for example, a couple of microseconds) for a link connecting two routers on the same university campus; however, dprop is hundreds of milliseconds for two routers interconnected by a geostationary satellite link, and can be
the dominant term in dnodal. Similarly, dtrans can range from negligible to significant.
Its contribution is typically negligible for transmission rates of 10 Mbps and higher
(for example, for LANs); however, it can be hundreds of milliseconds for large
Internet packets sent over low-speed dial-up modem links. The processing delay,
dproc, is often negligible; however, it strongly influences a router’s maximum
throughput, which is the maximum rate at which a router can forward packets.
1.4.2 Queuing Delay and Packet Loss
The most complicated and interesting component of nodal delay is the queuing
delay, dqueue. In fact, queuing delay is so important and interesting in computer networking that thousands of papers and numerous books have been written about it
[Bertsekas 1991; Daigle 1991; Kleinrock 1975, 1976; Ross 1995]. We give only a
high-level, intuitive discussion of queuing delay here; the more curious reader may
want to browse through some of the books (or even eventually write a PhD thesis
on the subject!). Unlike the other three delays (namely, dproc, dtrans, and dprop), the
queuing delay can vary from packet to packet. For example, if 10 packets arrive
at an empty queue at the same time, the first packet transmitted will suffer no queuing delay, while the last packet transmitted will suffer a relatively large queuing
delay (while it waits for the other nine packets to be transmitted). Therefore, when
39
40
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
characterizing queuing delay, one typically uses statistical measures, such as average queuing delay, variance of queuing delay, and the probability that the queuing
delay exceeds some specified value.
When is the queuing delay large and when is it insignificant? The answer to this
question depends on the rate at which traffic arrives at the queue, the transmission
rate of the link, and the nature of the arriving traffic, that is, whether the traffic arrives
periodically or arrives in bursts. To gain some insight here, let a denote the average
rate at which packets arrive at the queue (a is in units of packets/sec). Recall that R is
the transmission rate; that is, it is the rate (in bits/sec) at which bits are pushed out of
the queue. Also suppose, for simplicity, that all packets consist of L bits. Then the
average rate at which bits arrive at the queue is La bits/sec. Finally, assume that the
queue is very big, so that it can hold essentially an infinite number of bits. The ratio
La/R, called the traffic intensity, often plays an important role in estimating the
extent of the queuing delay. If La/R > 1, then the average rate at which bits arrive at
the queue exceeds the rate at which the bits can be transmitted from the queue. In this
unfortunate situation, the queue will tend to increase without bound and the queuing
delay will approach infinity! Therefore, one of the golden rules in traffic engineering
is: Design your system so that the traffic intensity is no greater than 1.
Now consider the case La/R ≤ 1. Here, the nature of the arriving traffic impacts
the queuing delay. For example, if packets arrive periodically—that is, one packet
arrives every L/R seconds—then every packet will arrive at an empty queue and
there will be no queuing delay. On the other hand, if packets arrive in bursts but
periodically, there can be a significant average queuing delay. For example, suppose
N packets arrive simultaneously every (L/R)N seconds. Then the first packet transmitted has no queuing delay; the second packet transmitted has a queuing delay of
L/R seconds; and more generally, the nth packet transmitted has a queuing delay of
(n ⫺ 1)L/R seconds. We leave it as an exercise for you to calculate the average queuing delay in this example.
The two examples of periodic arrivals described above are a bit academic. Typically, the arrival process to a queue is random; that is, the arrivals do not follow any
pattern and the packets are spaced apart by random amounts of time. In this more
realistic case, the quantity La/R is not usually sufficient to fully characterize the
queueing delay statistics. Nonetheless, it is useful in gaining an intuitive understanding of the extent of the queuing delay. In particular, if the traffic intensity is close to
zero, then packet arrivals are few and far between and it is unlikely that an arriving
packet will find another packet in the queue. Hence, the average queuing delay will
be close to zero. On the other hand, when the traffic intensity is close to 1, there will
be intervals of time when the arrival rate exceeds the transmission capacity (due to
variations in packet arrival rate), and a queue will form during these periods of time;
when the arrival rate is less than the transmission capacity, the length of the queue
will shrink. Nonetheless, as the traffic intensity approaches 1, the average queue
length gets larger and larger. The qualitative dependence of average queuing delay
on the traffic intensity is shown in Figure 1.18.
1.4
DELAY, LOSS, AND THROUGHPUT IN PACKET-SWITCHED NETWORKS
Average queuing delay
•
1
La/R
Figure 1.18
Dependence of average queuing delay on traffic intensity
One important aspect of Figure 1.18 is the fact that as the traffic intensity
approaches 1, the average queuing delay increases rapidly. A small percentage
increase in the intensity will result in a much larger percentage-wise increase in
delay. Perhaps you have experienced this phenomenon on the highway. If you regularly drive on a road that is typically congested, the fact that the road is typically
congested means that its traffic intensity is close to 1. If some event causes an even
slightly larger-than-usual amount of traffic, the delays you experience can be huge.
To really get a good feel for what queuing delays are about, you are encouraged
once again to visit the companion Web site, which provides an interactive Java
applet for a queue. If you set the packet arrival rate high enough so that the traffic
intensity exceeds 1, you will see the queue slowly build up over time.
Packet Loss
In our discussions above, we have assumed that the queue is capable of holding an
infinite number of packets. In reality a queue preceding a link has finite capacity,
although the queuing capacity greatly depends on the router design and cost.
Because the queue capacity is finite, packet delays do not really approach infinity as
the traffic intensity approaches 1. Instead, a packet can arrive to find a full queue.
With no place to store such a packet, a router will drop that packet; that is, the
packet will be lost. This overflow at a queue can again be seen in the Java applet for
a queue when the traffic intensity is greater than 1.
From an end-system viewpoint, a packet loss will look like a packet having
been transmitted into the network core but never emerging from the network at the
destination. The fraction of lost packets increases as the traffic intensity increases.
Therefore, performance at a node is often measured not only in terms of delay, but
also in terms of the probability of packet loss. As we’ll discuss in the subsequent
41
42
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
chapters, a lost packet may be retransmitted on an end-to-end basis in order to
ensure that all data are eventually transferred from source to destination
1.4.3 End-to-End Delay
Our discussion up to this point has focused on the nodal delay, that is, the delay at a
single router. Let’s now consider the total delay from source to destination. To get a
handle on this concept, suppose there are N ⫺ 1 routers between the source host and
the destination host. Let’s also suppose for the moment that the network is uncongested (so that queuing delays are negligible), the processing delay at each router
and at the source host is dproc, the transmission rate out of each router and out of the
source host is R bits/sec, and the propagation on each link is dprop. The nodal delays
accumulate and give an end-to-end delay,
dend-end = N (dproc + dtrans + dprop)
(1.2)
where, once again, dtrans = L/R, where L is the packet size. Note that Equation 1.2 is a
generalization of Equation 1.1, which did not take into account processing and propagation delays. We leave it to you to generalize Equation 1.2 to the case of heterogeneous delays at the nodes and to the presence of an average queuing delay at each node.
Traceroute
VideoNote
Using Traceroute to
discover network
paths and measure
network delay
To get a hands-on feel for end-to-end delay in a computer network, we can make use
of the Traceroute program. Traceroute is a simple program that can run in any Internet host. When the user specifies a destination hostname, the program in the source
host sends multiple, special packets toward that destination. As these packets work
their way toward the destination, they pass through a series of routers. When a
router receives one of these special packets, it sends back to the source a short message that contains the name and address of the router.
More specifically, suppose there are N ⫺ 1 routers between the source and the
destination. Then the source will send N special packets into the network, with each
packet addressed to the ultimate destination. These N special packets are marked 1
through N, with the first packet marked 1 and the last packet marked N. When the nth
router receives the nth packet marked n, the router does not forward the packet
toward its destination, but instead sends a message back to the source. When the destination host receives the Nth packet, it too returns a message back to the source. The
source records the time that elapses between when it sends a packet and when it
receives the corresponding return message; it also records the name and address of
the router (or the destination host) that returns the message. In this manner, the source
can reconstruct the route taken by packets flowing from source to destination, and
the source can determine the round-trip delays to all the intervening routers. Traceroute actually repeats the experiment just described three times, so the source actually
sends 3 • N packets to the destination. RFC 1393 describes Traceroute in detail.
1.4
•
DELAY, LOSS, AND THROUGHPUT IN PACKET-SWITCHED NETWORKS
43
Here is an example of the output of the Traceroute program, where the route
was being traced from the source host gaia.cs.umass.edu (at the University of Massachusetts) to the host cis.poly.edu (at Polytechnic University in Brooklyn). The output has six columns: the first column is the n value described above, that is, the
number of the router along the route; the second column is the name of the router;
the third column is the address of the router (of the form xxx.xxx.xxx.xxx); the last
three columns are the round-trip delays for three experiments. If the source receives
fewer than three messages from any given router (due to packet loss in the network),
Traceroute places an asterisk just after the router number and reports fewer than
three round-trip times for that router.
1 cs-gw (128.119.240.254) 1.009 ms 0.899 ms 0.993 ms
2 128.119.3.154 (128.119.3.154) 0.931 ms 0.441 ms 0.651 ms
3 border4-rt-gi-1-3.gw.umass.edu (128.119.2.194) 1.032 ms 0.484 ms 0.451 ms
4 acr1-ge-2-1-0.Boston.cw.net (208.172.51.129) 10.006 ms 8.150 ms 8.460 ms
5 agr4-loopback.NewYork.cw.net (206.24.194.104) 12.272 ms 14.344 ms 13.267 ms
6 acr2-loopback.NewYork.cw.net (206.24.194.62) 13.225 ms 12.292 ms 12.148 ms
7 pos10-2.core2.NewYork1.Level3.net (209.244.160.133) 12.218 ms 11.823 ms 11.793 ms
8 gige9-1-52.hsipaccess1.NewYork1.Level3.net (64.159.17.39) 13.081 ms 11.556 ms 13.297 ms
9 p0-0.polyu.bbnplanet.net (4.25.109.122) 12.716 ms 13.052 ms 12.786 ms
10 cis.poly.edu (128.238.32.126) 14.080 ms 13.035 ms 12.802 ms
In the trace above there are nine routers between the source and the destination.
Most of these routers have a name, and all of them have addresses. For example, the
name of Router 3 is border4-rt-gi-1-3.gw.umass.edu and its address is
128.119.2.194. Looking at the data provided for this same router, we see that
in the first of the three trials the round-trip delay between the source and the router
was 1.03 msec. The round-trip delays for the subsequent two trials were 0.48 and
0.45 msec. These round-trip delays include all of the delays just discussed, including transmission delays, propagation delays, router processing delays, and queuing
delays. Because the queuing delay is varying with time, the round-trip delay of
packet n sent to a router n can sometimes be longer than the round-trip delay of
packet n+1 sent to router n+1. Indeed, we observe this phenomenon in the above
example: the delays to Router 6 are larger than the delays to Router 7!
Want to try out Traceroute for yourself? We highly recommended that you visit
http://www.traceroute.org, which provides a Web interface to an extensive list of
sources for route tracing. You choose a source and supply the hostname for any destination. The Traceroute program then does all the work. There are a number of free
software programs that provide a graphical interface to Traceroute; one of our
favorites is PingPlotter [PingPlotter 2012].
End System, Application, and Other Delays
In addition to processing, transmission, and propagation delays, there can be additional significant delays in the end systems. For example, an end system wanting to
44
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
transmit a packet into a shared medium (e.g., as in a WiFi or cable modem scenario)
may purposefully delay its transmission as part of its protocol for sharing the
medium with other end systems; we’ll consider such protocols in detail in Chapter
5. Another important delay is media packetization delay, which is present in Voiceover-IP (VoIP) applications. In VoIP, the sending side must first fill a packet with
encoded digitized speech before passing the packet to the Internet. This time to fill a
packet—called the packetization delay—can be significant and can impact the userperceived quality of a VoIP call. This issue will be further explored in a homework
problem at the end of this chapter.
1.4.4 Throughput in Computer Networks
In addition to delay and packet loss, another critical performance measure in computer networks is end-to-end throughput. To define throughput, consider transferring a large file from Host A to Host B across a computer network. This transfer
might be, for example, a large video clip from one peer to another in a P2P file
sharing system. The instantaneous throughput at any instant of time is the rate
(in bits/sec) at which Host B is receiving the file. (Many applications, including
many P2P file sharing systems, display the instantaneous throughput during
downloads in the user interface—perhaps you have observed this before!) If the
file consists of F bits and the transfer takes T seconds for Host B to receive all F
bits, then the average throughput of the file transfer is F/T bits/sec. For some
applications, such as Internet telephony, it is desirable to have a low delay and an
instantaneous throughput consistently above some threshold (for example, over
24 kbps for some Internet telephony applications and over 256 kbps for some realtime video applications). For other applications, including those involving file
transfers, delay is not critical, but it is desirable to have the highest possible
throughput.
To gain further insight into the important concept of throughput, let’s consider a
few examples. Figure 1.19(a) shows two end systems, a server and a client, connected by two communication links and a router. Consider the throughput for a file
transfer from the server to the client. Let Rs denote the rate of the link between
the server and the router; and Rc denote the rate of the link between the router and
the client. Suppose that the only bits being sent in the entire network are those
from the server to the client. We now ask, in this ideal scenario, what is the serverto-client throughput? To answer this question, we may think of bits as fluid and
communication links as pipes. Clearly, the server cannot pump bits through its link
at a rate faster than Rs bps; and the router cannot forward bits at a rate faster than Rc
bps. If Rs < Rc, then the bits pumped by the server will “flow” right through the
router and arrive at the client at a rate of Rs bps, giving a throughput of Rs bps. If, on
the other hand, Rc < Rs, then the router will not be able to forward bits as quickly as
it receives them. In this case, bits will only leave the router at rate Rc, giving an
1.4
•
DELAY, LOSS, AND THROUGHPUT IN PACKET-SWITCHED NETWORKS
Rc
Rs
Server
a.
Client
R1
R2
Server
b.
Figure 1.19
RN
Client
Throughput for a file transfer from server to client
end-to-end throughput of Rc. (Note also that if bits continue to arrive at the router at
rate Rs, and continue to leave the router at Rc, the backlog of bits at the router waiting for transmission to the client will grow and grow—a most undesirable situation!)
Thus, for this simple two-link network, the throughput is min{Rc, Rs}, that is, it is
the transmission rate of the bottleneck link. Having determined the throughput, we
can now approximate the time it takes to transfer a large file of F bits from server to
client as F/min{Rs, Rc}. For a specific example, suppose you are downloading an
MP3 file of F = 32 million bits, the server has a transmission rate of Rs = 2 Mbps,
and you have an access link of Rc = 1 Mbps. The time needed to transfer the file is
then 32 seconds. Of course, these expressions for throughput and transfer time are
only approximations, as they do not account for store-and-forward and processing
delays as well as protocol issues.
Figure 1.19(b) now shows a network with N links between the server and the
client, with the transmission rates of the N links being R1, R2,..., RN. Applying the
same analysis as for the two-link network, we find that the throughput for a file
transfer from server to client is min{R1, R2,..., RN}, which is once again the transmission rate of the bottleneck link along the path between server and client.
Now consider another example motivated by today’s Internet. Figure 1.20(a)
shows two end systems, a server and a client, connected to a computer network.
Consider the throughput for a file transfer from the server to the client. The server is
connected to the network with an access link of rate Rs and the client is connected to
the network with an access link of rate Rc. Now suppose that all the links in the core
of the communication network have very high transmission rates, much higher than
Rs and Rc. Indeed, today, the core of the Internet is over-provisioned with high speed
links that experience little congestion. Also suppose that the only bits being sent in
the entire network are those from the server to the client. Because the core of the
computer network is like a wide pipe in this example, the rate at which bits can flow
45
46
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
Server
10 Servers
Rs
Bottleneck
link of
capacity R
Rc
Client
10 Clients
a.
Figure 1.20
b.
End-to-end throughput: (a) Client downloads a file from
server; (b) 10 clients downloading with 10 servers
from source to destination is again the minimum of Rs and Rc, that is, throughput =
min{Rs, Rc}. Therefore, the constraining factor for throughput in today’s Internet is
typically the access network.
For a final example, consider Figure 1.20(b) in which there are 10 servers and
10 clients connected to the core of the computer network. In this example, there are
10 simultaneous downloads taking place, involving 10 client-server pairs. Suppose
that these 10 downloads are the only traffic in the network at the current time. As
shown in the figure, there is a link in the core that is traversed by all 10 downloads.
Denote R for the transmission rate of this link R. Let’s suppose that all server access
links have the same rate Rs, all client access links have the same rate Rc, and the
transmission rates of all the links in the core—except the one common link of rate
R—are much larger than Rs, Rc, and R. Now we ask, what are the throughputs of the
downloads? Clearly, if the rate of the common link, R, is large—say a hundred times
larger than both Rs and Rc—then the throughput for each download will once again
be min{Rs, Rc}. But what if the rate of the common link is of the same order as Rs
and Rc? What will the throughput be in this case? Let’s take a look at a specific
1.5
•
PROTOCOL LAYERS AND THEIR SERVICE MODELS
example. Suppose Rs = 2 Mbps, Rc = 1 Mbps, R = 5 Mbps, and the common link
divides its transmission rate equally among the 10 downloads. Then the bottleneck
for each download is no longer in the access network, but is now instead the shared
link in the core, which only provides each download with 500 kbps of throughput.
Thus the end-to-end throughput for each download is now reduced to 500 kbps.
The examples in Figure 1.19 and Figure 1.20(a) show that throughput depends
on the transmission rates of the links over which the data flows. We saw that when
there is no other intervening traffic, the throughput can simply be approximated as
the minimum transmission rate along the path between source and destination. The
example in Figure 1.20(b) shows that more generally the throughput depends not
only on the transmission rates of the links along the path, but also on the intervening
traffic. In particular, a link with a high transmission rate may nonetheless be the bottleneck link for a file transfer if many other data flows are also passing through that
link. We will examine throughput in computer networks more closely in the homework problems and in the subsequent chapters.
1.5 Protocol Layers and Their Service Models
From our discussion thus far, it is apparent that the Internet is an extremely complicated system. We have seen that there are many pieces to the Internet: numerous
applications and protocols, various types of end systems, packet switches, and various types of link-level media. Given this enormous complexity, is there any hope of
organizing a network architecture, or at least our discussion of network architecture?
Fortunately, the answer to both questions is yes.
1.5.1 Layered Architecture
Before attempting to organize our thoughts on Internet architecture, let’s look for a
human analogy. Actually, we deal with complex systems all the time in our everyday life. Imagine if someone asked you to describe, for example, the airline system.
How would you find the structure to describe this complex system that has ticketing
agents, baggage checkers, gate personnel, pilots, airplanes, air traffic control, and a
worldwide system for routing airplanes? One way to describe this system might be
to describe the series of actions you take (or others take for you) when you fly on an
airline. You purchase your ticket, check your bags, go to the gate, and eventually get
loaded onto the plane. The plane takes off and is routed to its destination. After your
plane lands, you deplane at the gate and claim your bags. If the trip was bad, you
complain about the flight to the ticket agent (getting nothing for your effort). This
scenario is shown in Figure 1.21.
Already, we can see some analogies here with computer networking: You are
being shipped from source to destination by the airline; a packet is shipped from
47
48
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
Ticket (purchase)
Ticket (complain)
Baggage (check)
Baggage (claim)
Gates (load)
Gates (unload)
Runway takeoff
Runway landing
Airplane routing
Airplane routing
Airplane routing
Figure 1.21
Taking an airplane trip: actions
source host to destination host in the Internet. But this is not quite the analogy we
are after. We are looking for some structure in Figure 1.21. Looking at Figure 1.21,
we note that there is a ticketing function at each end; there is also a baggage function for already-ticketed passengers, and a gate function for already-ticketed and
already-baggage-checked passengers. For passengers who have made it through the
gate (that is, passengers who are already ticketed, baggage-checked, and through the
gate), there is a takeoff and landing function, and while in flight, there is an airplanerouting function. This suggests that we can look at the functionality in Figure 1.21
in a horizontal manner, as shown in Figure 1.22.
Figure 1.22 has divided the airline functionality into layers, providing a framework in which we can discuss airline travel. Note that each layer, combined with the
Ticket (purchase)
Ticket (complain) Ticket
Baggage (check)
Baggage (claim)
Gates (load)
Gates (unload)
Runway takeoff
Runway landing
Airplane routing
Departure airport
Airplane routing
Airplane routing
Baggage
Gate
Takeoff/Landing
Airplane routing
Intermediate air-traffic
control centers
Figure 1.22
Horizontal layering of airline functionality
1.5
•
PROTOCOL LAYERS AND THEIR SERVICE MODELS
layers below it, implements some functionality, some service. At the ticketing layer
and below, airline-counter-to-airline-counter transfer of a person is accomplished.
At the baggage layer and below, baggage-check-to-baggage-claim transfer of a person and bags is accomplished. Note that the baggage layer provides this service only
to an already-ticketed person. At the gate layer, departure-gate-to-arrival-gate transfer of a person and bags is accomplished. At the takeoff/landing layer, runway-torunway transfer of people and their bags is accomplished. Each layer provides its
service by (1) performing certain actions within that layer (for example, at the gate
layer, loading and unloading people from an airplane) and by (2) using the services
of the layer directly below it (for example, in the gate layer, using the runway-torunway passenger transfer service of the takeoff/landing layer).
A layered architecture allows us to discuss a well-defined, specific part of a
large and complex system. This simplification itself is of considerable value by providing modularity, making it much easier to change the implementation of the service provided by the layer. As long as the layer provides the same service to the layer
above it, and uses the same services from the layer below it, the remainder of the
system remains unchanged when a layer’s implementation is changed. (Note that
changing the implementation of a service is very different from changing the service itself!) For example, if the gate functions were changed (for instance, to have
people board and disembark by height), the remainder of the airline system would
remain unchanged since the gate layer still provides the same function (loading and
unloading people); it simply implements that function in a different manner after the
change. For large and complex systems that are constantly being updated, the ability
to change the implementation of a service without affecting other components of the
system is another important advantage of layering.
Protocol Layering
But enough about airlines. Let’s now turn our attention to network protocols. To
provide structure to the design of network protocols, network designers organize
protocols—and the network hardware and software that implement the protocols—
in layers. Each protocol belongs to one of the layers, just as each function in the
airline architecture in Figure 1.22 belonged to a layer. We are again interested in
the services that a layer offers to the layer above—the so-called service model of
a layer. Just as in the case of our airline example, each layer provides its service
by (1) performing certain actions within that layer and by (2) using the services of
the layer directly below it. For example, the services provided by layer n may
include reliable delivery of messages from one edge of the network to the other.
This might be implemented by using an unreliable edge-to-edge message delivery
service of layer n ⫺ 1, and adding layer n functionality to detect and retransmit
lost messages.
A protocol layer can be implemented in software, in hardware, or in a combination of the two. Application-layer protocols—such as HTTP and SMTP—are almost
49
50
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
Application
Presentation
Application
Session
Transport
Transport
Network
Network
Link
Link
Physical
Physical
a. Five-layer
Internet
protocol stack
Figure 1.23
b. Seven-layer
ISO OSI
reference model
The Internet protocol stack (a) and OSI reference model (b)
always implemented in software in the end systems; so are transport-layer protocols.
Because the physical layer and data link layers are responsible for handling communication over a specific link, they are typically implemented in a network interface
card (for example, Ethernet or WiFi interface cards) associated with a given link.
The network layer is often a mixed implementation of hardware and software. Also
note that just as the functions in the layered airline architecture were distributed
among the various airports and flight control centers that make up the system, so too
is a layer n protocol distributed among the end systems, packet switches, and other
components that make up the network. That is, there’s often a piece of a layer n protocol in each of these network components.
Protocol layering has conceptual and structural advantages [RFC 3439]. As we
have seen, layering provides a structured way to discuss system components. Modularity makes it easier to update system components. We mention, however, that
some researchers and networking engineers are vehemently opposed to layering
[Wakeman 1992]. One potential drawback of layering is that one layer may duplicate lower-layer functionality. For example, many protocol stacks provide error
recovery on both a per-link basis and an end-to-end basis. A second potential drawback is that functionality at one layer may need information (for example, a timestamp value) that is present only in another layer; this violates the goal of
separation of layers.
When taken together, the protocols of the various layers are called the protocol
stack. The Internet protocol stack consists of five layers: the physical, link, network,
transport, and application layers, as shown in Figure 1.23(a). If you examine the
Table of Contents, you will see that we have roughly organized this book using the
layers of the Internet protocol stack. We take a top-down approach, first covering
the application layer and then proceeding downward.
1.5
•
PROTOCOL LAYERS AND THEIR SERVICE MODELS
Application Layer
The application layer is where network applications and their application-layer protocols reside. The Internet’s application layer includes many protocols, such as the HTTP
protocol (which provides for Web document request and transfer), SMTP (which provides for the transfer of e-mail messages), and FTP (which provides for the transfer of
files between two end systems). We’ll see that certain network functions, such as the
translation of human-friendly names for Internet end systems like www.ietf.org to a
32-bit network address, are also done with the help of a specific application-layer protocol, namely, the domain name system (DNS). We’ll see in Chapter 2 that it is very
easy to create and deploy our own new application-layer protocols.
An application-layer protocol is distributed over multiple end systems, with the
application in one end system using the protocol to exchange packets of information
with the application in another end system. We’ll refer to this packet of information
at the application layer as a message.
Transport Layer
The Internet’s transport layer transports application-layer messages between
application endpoints. In the Internet there are two transport protocols, TCP and
UDP, either of which can transport application-layer messages. TCP provides a
connection-oriented service to its applications. This service includes guaranteed
delivery of application-layer messages to the destination and flow control (that is,
sender/receiver speed matching). TCP also breaks long messages into shorter segments and provides a congestion-control mechanism, so that a source throttles its
transmission rate when the network is congested. The UDP protocol provides a connectionless service to its applications. This is a no-frills service that provides no
reliability, no flow control, and no congestion control. In this book, we’ll refer to a
transport-layer packet as a segment.
Network Layer
The Internet’s network layer is responsible for moving network-layer packets
known as datagrams from one host to another. The Internet transport-layer protocol (TCP or UDP) in a source host passes a transport-layer segment and a destination address to the network layer, just as you would give the postal service a letter
with a destination address. The network layer then provides the service of delivering the segment to the transport layer in the destination host.
The Internet’s network layer includes the celebrated IP Protocol, which defines
the fields in the datagram as well as how the end systems and routers act on these
fields. There is only one IP protocol, and all Internet components that have a network layer must run the IP protocol. The Internet’s network layer also contains routing protocols that determine the routes that datagrams take between sources and
51
52
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
destinations. The Internet has many routing protocols. As we saw in Section 1.3, the
Internet is a network of networks, and within a network, the network administrator
can run any routing protocol desired. Although the network layer contains both the
IP protocol and numerous routing protocols, it is often simply referred to as the IP
layer, reflecting the fact that IP is the glue that binds the Internet together.
Link Layer
The Internet’s network layer routes a datagram through a series of routers between
the source and destination. To move a packet from one node (host or router) to the
next node in the route, the network layer relies on the services of the link layer. In
particular, at each node, the network layer passes the datagram down to the link
layer, which delivers the datagram to the next node along the route. At this next
node, the link layer passes the datagram up to the network layer.
The services provided by the link layer depend on the specific link-layer protocol that is employed over the link. For example, some link-layer protocols provide
reliable delivery, from transmitting node, over one link, to receiving node. Note that
this reliable delivery service is different from the reliable delivery service of TCP,
which provides reliable delivery from one end system to another. Examples of linklayer protocols include Ethernet, WiFi, and the cable access network’s DOCSIS protocol. As datagrams typically need to traverse several links to travel from source to
destination, a datagram may be handled by different link-layer protocols at different
links along its route. For example, a datagram may be handled by Ethernet on one
link and by PPP on the next link. The network layer will receive a different service
from each of the different link-layer protocols. In this book, we’ll refer to the linklayer packets as frames.
Physical Layer
While the job of the link layer is to move entire frames from one network element
to an adjacent network element, the job of the physical layer is to move the individual bits within the frame from one node to the next. The protocols in this layer are
again link dependent and further depend on the actual transmission medium of the
link (for example, twisted-pair copper wire, single-mode fiber optics). For example,
Ethernet has many physical-layer protocols: one for twisted-pair copper wire,
another for coaxial cable, another for fiber, and so on. In each case, a bit is moved
across the link in a different way.
The OSI Model
Having discussed the Internet protocol stack in detail, we should mention that it is
not the only protocol stack around. In particular, back in the late 1970s, the International Organization for Standardization (ISO) proposed that computer networks be
1.5
•
PROTOCOL LAYERS AND THEIR SERVICE MODELS
organized around seven layers, called the Open Systems Interconnection (OSI)
model [ISO 2012]. The OSI model took shape when the protocols that were to
become the Internet protocols were in their infancy, and were but one of many different protocol suites under development; in fact, the inventors of the original OSI
model probably did not have the Internet in mind when creating it. Nevertheless,
beginning in the late 1970s, many training and university courses picked up on the
ISO mandate and organized courses around the seven-layer model. Because of its
early impact on networking education, the seven-layer model continues to linger on
in some networking textbooks and training courses.
The seven layers of the OSI reference model, shown in Figure 1.23(b), are:
application layer, presentation layer, session layer, transport layer, network layer,
data link layer, and physical layer. The functionality of five of these layers is
roughly the same as their similarly named Internet counterparts. Thus, let’s consider
the two additional layers present in the OSI reference model—the presentation layer
and the session layer. The role of the presentation layer is to provide services that
allow communicating applications to interpret the meaning of data exchanged.
These services include data compression and data encryption (which are selfexplanatory) as well as data description (which, as we will see in Chapter 9, frees
the applications from having to worry about the internal format in which data are
represented/stored—formats that may differ from one computer to another). The
session layer provides for delimiting and synchronization of data exchange, including the means to build a checkpointing and recovery scheme.
The fact that the Internet lacks two layers found in the OSI reference model
poses a couple of interesting questions: Are the services provided by these layers
unimportant? What if an application needs one of these services? The Internet’s
answer to both of these questions is the same—it’s up to the application developer.
It’s up to the application developer to decide if a service is important, and if the
service is important, it’s up to the application developer to build that functionality
into the application.
1.5.2 Encapsulation
Figure 1.24 shows the physical path that data takes down a sending end system’s
protocol stack, up and down the protocol stacks of an intervening link-layer switch
and router, and then up the protocol stack at the receiving end system. As we discuss
later in this book, routers and link-layer switches are both packet switches. Similar
to end systems, routers and link-layer switches organize their networking hardware
and software into layers. But routers and link-layer switches do not implement all of
the layers in the protocol stack; they typically implement only the bottom layers. As
shown in Figure 1.24, link-layer switches implement layers 1 and 2; routers implement layers 1 through 3. This means, for example, that Internet routers are capable
of implementing the IP protocol (a layer 3 protocol), while link-layer switches are
not. We’ll see later that while link-layer switches do not recognize IP addresses, they
53
54
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
Source
Message
M
Segment Ht M
Datagram Hn Ht M
Frame
H l Hn Ht M
Application
Transport
Network
Link
Physical
H l Hn Ht M
Link
H l Hn Ht M
Physical
Link-layer switch
Router
Destination
M
Ht M
Hn Ht M
H l Hn Ht M
Application
Transport
Hn H t M
Network
H l Hn H t M
Link
Network
Link
Hn Ht M
H l Hn Ht M
Physical
Physical
Figure 1.24
Hosts, routers, and link-layer switches; each contains a
different set of layers, reflecting their differences in
functionality
are capable of recognizing layer 2 addresses, such as Ethernet addresses. Note that
hosts implement all five layers; this is consistent with the view that the Internet
architecture puts much of its complexity at the edges of the network.
Figure 1.24 also illustrates the important concept of encapsulation. At the
sending host, an application-layer message (M in Figure 1.24) is passed to the
transport layer. In the simplest case, the transport layer takes the message and
appends additional information (so-called transport-layer header information, Ht
in Figure 1.24) that will be used by the receiver-side transport layer. The application-layer message and the transport-layer header information together constitute
the transport-layer segment. The transport-layer segment thus encapsulates the
application-layer message. The added information might include information
allowing the receiver-side transport layer to deliver the message up to the appropriate application, and error-detection bits that allow the receiver to determine
whether bits in the message have been changed in route. The transport layer then
passes the segment to the network layer, which adds network-layer header information (Hn in Figure 1.24) such as source and destination end system addresses,
1.6
•
NETWORKS UNDER ATTACK
creating a network-layer datagram. The datagram is then passed to the link
layer, which (of course!) will add its own link-layer header information and create
a link-layer frame. Thus, we see that at each layer, a packet has two types of
fields: header fields and a payload field. The payload is typically a packet from
the layer above.
A useful analogy here is the sending of an interoffice memo from one corporate branch office to another via the public postal service. Suppose Alice, who is
in one branch office, wants to send a memo to Bob, who is in another branch
office. The memo is analogous to the application-layer message. Alice puts the
memo in an interoffice envelope with Bob’s name and department written on the
front of the envelope. The interoffice envelope is analogous to a transport-layer
segment—it contains header information (Bob’s name and department number)
and it encapsulates the application-layer message (the memo). When the sending
branch-office mailroom receives the interoffice envelope, it puts the interoffice
envelope inside yet another envelope, which is suitable for sending through the
public postal service. The sending mailroom also writes the postal address of the
sending and receiving branch offices on the postal envelope. Here, the postal
envelope is analogous to the datagram—it encapsulates the transport-layer segment (the interoffice envelope), which encapsulates the original message (the
memo). The postal service delivers the postal envelope to the receiving branchoffice mailroom. There, the process of de-encapsulation is begun. The mailroom
extracts the interoffice memo and forwards it to Bob. Finally, Bob opens the envelope and removes the memo.
The process of encapsulation can be more complex than that described above.
For example, a large message may be divided into multiple transport-layer segments
(which might themselves each be divided into multiple network-layer datagrams).
At the receiving end, such a segment must then be reconstructed from its constituent
datagrams.
1.6 Networks Under Attack
The Internet has become mission critical for many institutions today, including large
and small companies, universities, and government agencies. Many individuals also
rely on the Internet for many of their professional, social, and personal activities.
But behind all this utility and excitement, there is a dark side, a side where “bad
guys” attempt to wreak havoc in our daily lives by damaging our Internet-connected
computers, violating our privacy, and rendering inoperable the Internet services on
which we depend.
The field of network security is about how the bad guys can attack computer
networks and about how we, soon-to-be experts in computer networking, can
55
56
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
defend networks against those attacks, or better yet, design new architectures that
are immune to such attacks in the first place. Given the frequency and variety of
existing attacks as well as the threat of new and more destructive future attacks,
network security has become a central topic in the field of computer networking.
One of the features of this textbook is that it brings network security issues to the
forefront.
Since we don’t yet have expertise in computer networking and Internet protocols, we’ll begin here by surveying some of today’s more prevalent securityrelated problems. This will whet our appetite for more substantial discussions in
the upcoming chapters. So we begin here by simply asking, what can go wrong?
How are computer networks vulnerable? What are some of the more prevalent
types of attacks today?
The bad guys can put malware into your host via the Internet
We attach devices to the Internet because we want to receive/send data from/to
the Internet. This includes all kinds of good stuff, including Web pages, e-mail
messages, MP3s, telephone calls, live video, search engine results, and so on.
But, unfortunately, along with all that good stuff comes malicious stuff—collectively known as malware—that can also enter and infect our devices. Once malware infects our device it can do all kinds of devious things, including deleting
our files; installing spyware that collects our private information, such as social
security numbers, passwords, and keystrokes, and then sends this (over the Internet, of course!) back to the bad guys. Our compromised host may also be
enrolled in a network of thousands of similarly compromised devices, collectively known as a botnet, which the bad guys control and leverage for spam email distribution or distributed denial-of-service attacks (soon to be discussed)
against targeted hosts.
Much of the malware out there today is self-replicating: once it infects one
host, from that host it seeks entry into other hosts over the Internet, and from the
newly infected hosts, it seeks entry into yet more hosts. In this manner, selfreplicating malware can spread exponentially fast. Malware can spread in the
form of a virus or a worm. Viruses are malware that require some form of user
interaction to infect the user’s device. The classic example is an e-mail attachment
containing malicious executable code. If a user receives and opens such an attachment, the user inadvertently runs the malware on the device. Typically, such email viruses are self-replicating: once executed, the virus may send an identical
message with an identical malicious attachment to, for example, every recipient in
the user’s address book. Worms are malware that can enter a device without any
explicit user interaction. For example, a user may be running a vulnerable network
application to which an attacker can send malware. In some cases, without any
user intervention, the application may accept the malware from the Internet and
1.6
•
NETWORKS UNDER ATTACK
run it, creating a worm. The worm in the newly infected device then scans the
Internet, searching for other hosts running the same vulnerable network application. When it finds other vulnerable hosts, it sends a copy of itself to those hosts.
Today, malware, is pervasive and costly to defend against. As you work through
this textbook, we encourage you to think about the following question: What can
computer network designers do to defend Internet-attached devices from malware
attacks?
The bad guys can attack servers and network infrastructure
Another broad class of security threats are known as denial-of-service (DoS)
attacks. As the name suggests, a DoS attack renders a network, host, or other piece
of infrastructure unusable by legitimate users. Web servers, e-mail servers, DNS
servers (discussed in Chapter 2), and institutional networks can all be subject to DoS
attacks. Internet DoS attacks are extremely common, with thousands of DoS attacks
occurring every year [Moore 2001; Mirkovic 2005]. Most Internet DoS attacks fall
into one of three categories:
• Vulnerability attack. This involves sending a few well-crafted messages to a vulnerable application or operating system running on a targeted host. If the right
sequence of packets is sent to a vulnerable application or operating system, the
service can stop or, worse, the host can crash.
• Bandwidth flooding. The attacker sends a deluge of packets to the targeted
host—so many packets that the target’s access link becomes clogged, preventing
legitimate packets from reaching the server.
• Connection flooding. The attacker establishes a large number of half-open or
fully open TCP connections (TCP connections are discussed in Chapter 3) at the
target host. The host can become so bogged down with these bogus connections
that it stops accepting legitimate connections.
Let’s now explore the bandwidth-flooding attack in more detail. Recalling our
delay and loss analysis discussion in Section 1.4.2, it’s evident that if the server has an
access rate of R bps, then the attacker will need to send traffic at a rate of approximately R bps to cause damage. If R is very large, a single attack source may not be
able to generate enough traffic to harm the server. Furthermore, if all the traffic
emanates from a single source, an upstream router may be able to detect the attack and
block all traffic from that source before the traffic gets near the server. In a distributed
DoS (DDoS) attack, illustrated in Figure 1.25, the attacker controls multiple sources
and has each source blast traffic at the target. With this approach, the aggregate traffic
rate across all the controlled sources needs to be approximately R to cripple the
57
58
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
Slave
Slave
“start
attack”
Victim
Slave
Attacker
Slave
Slave
Figure 1.25
A distributed denial-of-service attack
service. DDoS attacks leveraging botnets with thousands of comprised hosts are a
common occurrence today [Mirkovic 2005]. DDos attacks are much harder to detect
and defend against than a DoS attack from a single host.
We encourage you to consider the following question as you work your way
through this book: What can computer network designers do to defend against DoS
attacks? We will see that different defenses are needed for the three types of DoS
attacks.
The bad guys can sniff packets
Many users today access the Internet via wireless devices, such as WiFi-connected
laptops or handheld devices with cellular Internet connections (covered in Chapter
6). While ubiquitous Internet access is extremely convenient and enables marvelous
new applications for mobile users, it also creates a major security vulnerability—by
placing a passive receiver in the vicinity of the wireless transmitter, that receiver can
obtain a copy of every packet that is transmitted! These packets can contain all kinds
of sensitive information, including passwords, social security numbers, trade
secrets, and private personal messages. A passive receiver that records a copy of
every packet that flies by is called a packet sniffer.
Sniffers can be deployed in wired environments as well. In wired broadcast environments, as in many Ethernet LANs, a packet sniffer can obtain copies of broadcast
packets sent over the LAN. As described in Section 1.2, cable access technologies
also broadcast packets and are thus vulnerable to sniffing. Furthermore, a bad guy
who gains access to an institution’s access router or access link to the Internet may
1.6
•
NETWORKS UNDER ATTACK
be able to plant a sniffer that makes a copy of every packet going to/from the organization. Sniffed packets can then be analyzed offline for sensitive information.
Packet-sniffing software is freely available at various Web sites and as commercial
products. Professors teaching a networking course have been known to assign lab exercises that involve writing a packet-sniffing and application-layer data reconstruction
program. Indeed, the Wireshark [Wireshark 2012] labs associated with this text (see the
introductory Wireshark lab at the end of this chapter) use exactly such a packet sniffer!
Because packet sniffers are passive—that is, they do not inject packets into the
channel—they are difficult to detect. So, when we send packets into a wireless channel, we must accept the possibility that some bad guy may be recording copies of
our packets. As you may have guessed, some of the best defenses against packet
sniffing involve cryptography. We will examine cryptography as it applies to network security in Chapter 8.
The bad guys can masquerade as someone you trust
It is surprisingly easy (you will have the knowledge to do so shortly as you proceed
through this text!) to create a packet with an arbitrary source address, packet content, and destination address and then transmit this hand-crafted packet into the
Internet, which will dutifully forward the packet to its destination. Imagine the
unsuspecting receiver (say an Internet router) who receives such a packet, takes the
(false) source address as being truthful, and then performs some command embedded in the packet’s contents (say modifies its forwarding table). The ability to inject
packets into the Internet with a false source address is known as IP spoofing, and is
but one of many ways in which one user can masquerade as another user.
To solve this problem, we will need end-point authentication, that is, a mechanism that will allow us to determine with certainty if a message originates from
where we think it does. Once again, we encourage you to think about how this can
be done for network applications and protocols as you progress through the chapters
of this book. We will explore mechanisms for end-point authentication in Chapter 8.
In closing this section, it’s worth considering how the Internet got to be such an
insecure place in the first place. The answer, in essence, is that the Internet was originally designed to be that way, based on the model of “a group of mutually trusting
users attached to a transparent network” [Blumenthal 2001]—a model in which (by
definition) there is no need for security. Many aspects of the original Internet architecture deeply reflect this notion of mutual trust. For example, the ability for one
user to send a packet to any other user is the default rather than a requested/granted
capability, and user identity is taken at declared face value, rather than being authenticated by default.
But today’s Internet certainly does not involve “mutually trusting users.”
Nonetheless, today’s users still need to communicate when they don’t necessarily
trust each other, may wish to communicate anonymously, may communicate indirectly through third parties (e.g., Web caches, which we’ll study in Chapter 2, or
59
60
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
mobility-assisting agents, which we’ll study in Chapter 6), and may distrust the
hardware, software, and even the air through which they communicate. We now
have many security-related challenges before us as we progress through this book:
We should seek defenses against sniffing, end-point masquerading, man-in-themiddle attacks, DDoS attacks, malware, and more. We should keep in mind that
communication among mutually trusted users is the exception rather than the rule.
Welcome to the world of modern computer networking!
1.7 History of Computer Networking and
the Internet
Sections 1.1 through 1.6 presented an overview of the technology of computer networking and the Internet. You should know enough now to impress your family and
friends! However, if you really want to be a big hit at the next cocktail party, you
should sprinkle your discourse with tidbits about the fascinating history of the Internet [Segaller 1998].
1.7.1 The Development of Packet Switching: 1961–1972
The field of computer networking and today’s Internet trace their beginnings
back to the early 1960s, when the telephone network was the world’s dominant
communication network. Recall from Section 1.3 that the telephone network uses
circuit switching to transmit information from a sender to a receiver—an appropriate choice given that voice is transmitted at a constant rate between sender
and receiver. Given the increasing importance of computers in the early 1960s
and the advent of timeshared computers, it was perhaps natural to consider how
to hook computers together so that they could be shared among geographically
distributed users. The traffic generated by such users was likely to be bursty—
intervals of activity, such as the sending of a command to a remote computer, followed by periods of inactivity while waiting for a reply or while contemplating
the received response.
Three research groups around the world, each unaware of the others’ work
[Leiner 1998], began inventing packet switching as an efficient and robust alternative to circuit switching. The first published work on packet-switching techniques
was that of Leonard Kleinrock [Kleinrock 1961; Kleinrock 1964], then a graduate
student at MIT. Using queuing theory, Kleinrock’s work elegantly demonstrated the
effectiveness of the packet-switching approach for bursty traffic sources. In 1964,
Paul Baran [Baran 1964] at the Rand Institute had begun investigating the use of
packet switching for secure voice over military networks, and at the National Physical Laboratory in England, Donald Davies and Roger Scantlebury were also developing their ideas on packet switching.
1.7
•
HISTORY OF COMPUTER NETWORKING AND THE INTERNET
The work at MIT, Rand, and the NPL laid the foundations for today’s Internet. But the Internet also has a long history of a let’s-build-it-and-demonstrate-it
attitude that also dates back to the 1960s. J. C. R. Licklider [DEC 1990] and
Lawrence Roberts, both colleagues of Kleinrock’s at MIT, went on to lead the
computer science program at the Advanced Research Projects Agency (ARPA) in
the United States. Roberts published an overall plan for the ARPAnet [Roberts
1967], the first packet-switched computer network and a direct ancestor of today’s
public Internet. On Labor Day in 1969, the first packet switch was installed at
UCLA under Kleinrock’s supervision, and three additional packet switches were
installed shortly thereafter at the Stanford Research Institute (SRI), UC Santa Barbara, and the University of Utah (Figure 1.26). The fledgling precursor to the
Figure 1.26
An early packet switch
61
62
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
Internet was four nodes large by the end of 1969. Kleinrock recalls the very first
use of the network to perform a remote login from UCLA to SRI, crashing the system [Kleinrock 2004].
By 1972, ARPAnet had grown to approximately 15 nodes and was given its first
public demonstration by Robert Kahn. The first host-to-host protocol between
ARPAnet end systems, known as the network-control protocol (NCP), was completed [RFC 001]. With an end-to-end protocol available, applications could now be
written. Ray Tomlinson wrote the first e-mail program in 1972.
1.7.2 Proprietary Networks and Internetworking: 1972–1980
The initial ARPAnet was a single, closed network. In order to communicate with an
ARPAnet host, one had to be actually attached to another ARPAnet IMP. In the early
to mid-1970s, additional stand-alone packet-switching networks besides ARPAnet
came into being: ALOHANet, a microwave network linking universities on the
Hawaiian islands [Abramson 1970], as well as DARPA’s packet-satellite [RFC 829]
and packet-radio networks [Kahn 1978]; Telenet, a BBN commercial packetswitching network based on ARPAnet technology; Cyclades, a French packetswitching network pioneered by Louis Pouzin [Think 2012]; Time-sharing networks
such as Tymnet and the GE Information Services network, among others, in the late
1960s and early 1970s [Schwartz 1977]; IBM’s SNA (1969–1974), which paralleled the ARPAnet work [Schwartz 1977].
The number of networks was growing. With perfect hindsight we can see that
the time was ripe for developing an encompassing architecture for connecting networks together. Pioneering work on interconnecting networks (under the sponsorship of the Defense Advanced Research Projects Agency (DARPA)), in essence
creating a network of networks, was done by Vinton Cerf and Robert Kahn [Cerf
1974]; the term internetting was coined to describe this work.
These architectural principles were embodied in TCP. The early versions of
TCP, however, were quite different from today’s TCP. The early versions of TCP
combined a reliable in-sequence delivery of data via end-system retransmission
(still part of today’s TCP) with forwarding functions (which today are performed
by IP). Early experimentation with TCP, combined with the recognition of the
importance of an unreliable, non-flow-controlled, end-to-end transport service
for applications such as packetized voice, led to the separation of IP out of TCP
and the development of the UDP protocol. The three key Internet protocols that
we see today—TCP, UDP, and IP—were conceptually in place by the end of the
1970s.
In addition to the DARPA Internet-related research, many other important
networking activities were underway. In Hawaii, Norman Abramson was developing ALOHAnet, a packet-based radio network that allowed multiple remote
sites on the Hawaiian Islands to communicate with each other. The ALOHA protocol
1.7
•
HISTORY OF COMPUTER NETWORKING AND THE INTERNET
[Abramson 1970] was the first multiple-access protocol, allowing geographically
distributed users to share a single broadcast communication medium (a radio frequency). Metcalfe and Boggs built on Abramson’s multiple-access protocol work
when they developed the Ethernet protocol [Metcalfe 1976] for wire-based
shared broadcast networks. Interestingly, Metcalfe and Boggs’ Ethernet protocol
was motivated by the need to connect multiple PCs, printers, and shared disks
[Perkins 1994]. Twenty-five years ago, well before the PC revolution and the
explosion of networks, Metcalfe and Boggs were laying the foundation for
today’s PC LANs.
1.7.3 A Proliferation of Networks: 1980–1990
By the end of the 1970s, approximately two hundred hosts were connected to the
ARPAnet. By the end of the 1980s the number of hosts connected to the public
Internet, a confederation of networks looking much like today’s Internet, would
reach a hundred thousand. The 1980s would be a time of tremendous growth.
Much of that growth resulted from several distinct efforts to create computer
networks linking universities together. BITNET provided e-mail and file transfers
among several universities in the Northeast. CSNET (computer science network)
was formed to link university researchers who did not have access to ARPAnet. In
1986, NSFNET was created to provide access to NSF-sponsored supercomputing
centers. Starting with an initial backbone speed of 56 kbps, NSFNET’s backbone
would be running at 1.5 Mbps by the end of the decade and would serve as a primary backbone linking regional networks.
In the ARPAnet community, many of the final pieces of today’s Internet architecture were falling into place. January 1, 1983 saw the official deployment of
TCP/IP as the new standard host protocol for ARPAnet (replacing the NCP protocol). The transition [RFC 801] from NCP to TCP/IP was a flag day event—all
hosts were required to transfer over to TCP/IP as of that day. In the late 1980s,
important extensions were made to TCP to implement host-based congestion control [Jacobson 1988]. The DNS, used to map between a human-readable Internet
name (for example, gaia.cs.umass.edu) and its 32-bit IP address, was also developed [RFC 1034].
Paralleling this development of the ARPAnet (which was for the most part a
US effort), in the early 1980s the French launched the Minitel project, an ambitious plan to bring data networking into everyone’s home. Sponsored by the
French government, the Minitel system consisted of a public packet-switched network (based on the X.25 protocol suite), Minitel servers, and inexpensive terminals with built-in low-speed modems. The Minitel became a huge success in 1984
when the French government gave away a free Minitel terminal to each French
household that wanted one. Minitel sites included free sites—such as a telephone
directory site—as well as private sites, which collected a usage-based fee from
63
64
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
each user. At its peak in the mid 1990s, it offered more than 20,000 services, ranging from home banking to specialized research databases. The Minitel was in a
large proportion of French homes 10 years before most Americans had ever heard
of the Internet.
1.7.4 The Internet Explosion: The 1990s
The 1990s were ushered in with a number of events that symbolized the continued
evolution and the soon-to-arrive commercialization of the Internet. ARPAnet, the
progenitor of the Internet, ceased to exist. In 1991, NSFNET lifted its restrictions on
the use of NSFNET for commercial purposes. NSFNET itself would be decommissioned in 1995, with Internet backbone traffic being carried by commercial Internet
Service Providers.
The main event of the 1990s was to be the emergence of the World Wide Web
application, which brought the Internet into the homes and businesses of millions of
people worldwide. The Web served as a platform for enabling and deploying hundreds of new applications that we take for granted today, including search (e.g.,
Google and Bing) Internet commerce (e.g., Amazon and eBay) and social networks
(e.g., Facebook).
The Web was invented at CERN by Tim Berners-Lee between 1989 and 1991
[Berners-Lee 1989], based on ideas originating in earlier work on hypertext from
the 1940s by Vannevar Bush [Bush 1945] and since the 1960s by Ted Nelson
[Xanadu 2012]. Berners-Lee and his associates developed initial versions of HTML,
HTTP, a Web server, and a browser—the four key components of the Web. Around
the end of 1993 there were about two hundred Web servers in operation, this collection of servers being just a harbinger of what was about to come. At about this time
several researchers were developing Web browsers with GUI interfaces, including
Marc Andreessen, who along with Jim Clark, formed Mosaic Communications,
which later became Netscape Communications Corporation [Cusumano 1998; Quittner
1998]. By 1995, university students were using Netscape browsers to surf the Web
on a daily basis. At about this time companies—big and small—began to operate
Web servers and transact commerce over the Web. In 1996, Microsoft started to
make browsers, which started the browser war between Netscape and Microsoft,
which Microsoft won a few years later [Cusumano 1998].
The second half of the 1990s was a period of tremendous growth and innovation for the Internet, with major corporations and thousands of startups creating
Internet products and services. By the end of the millennium the Internet was supporting hundreds of popular applications, including four killer applications:
• E-mail, including attachments and Web-accessible e-mail
• The Web, including Web browsing and Internet commerce
1.7
•
HISTORY OF COMPUTER NETWORKING AND THE INTERNET
• Instant messaging, with contact lists
• Peer-to-peer file sharing of MP3s, pioneered by Napster
Interestingly, the first two killer applications came from the research community,
whereas the last two were created by a few young entrepreneurs.
The period from 1995 to 2001 was a roller-coaster ride for the Internet in the
financial markets. Before they were even profitable, hundreds of Internet startups
made initial public offerings and started to be traded in a stock market. Many companies were valued in the billions of dollars without having any significant revenue
streams. The Internet stocks collapsed in 2000–2001, and many startups shut down.
Nevertheless, a number of companies emerged as big winners in the Internet space,
including Microsoft, Cisco, Yahoo, e-Bay, Google, and Amazon.
1.7.5 The New Millennium
Innovation in computer networking continues at a rapid pace. Advances are being
made on all fronts, including deployments of faster routers and higher transmission
speeds in both access networks and in network backbones. But the following developments merit special attention:
• Since the beginning of the millennium, we have been seeing aggressive deployment of broadband Internet access to homes—not only cable modems and DSL
but also fiber to the home, as discussed in Section 1.2. This high-speed Internet
access has set the stage for a wealth of video applications, including the distribution of user-generated video (for example, YouTube), on-demand streaming of
movies and television shows (e.g., Netflix) , and multi-person video conference
(e.g., Skype).
• The increasing ubiquity of high-speed (54 Mbps and higher) public WiFi networks and medium-speed (up to a few Mbps) Internet access via 3G and 4G
cellular telephony networks is not only making it possible to remain constantly connected while on the move, but also enabling new location-specific
applications. The number of wireless devices connecting to the Internet surpassed the number of wired devices in 2011. This high-speed wireless access
has set the stage for the rapid emergence of hand-held computers (iPhones,
Androids, iPads, and so on), which enjoy constant and untethered access to
the Internet.
• Online social networks, such as Facebook and Twitter, have created massive people networks on top of the Internet. Many Internet users today “live” primarily
within Facebook. Through their APIs, the online social networks create platforms for new networked applications and distributed games.
65
66
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
• As discussed in Section 1.3.3, online service providers, such as Google and
Microsoft, have deployed their own extensive private networks, which not only
connect together their globally distributed data centers, but are used to bypass
the Internet as much as possible by peering directly with lower-tier ISPs. As a
result, Google provides search results and email access almost instantaneously,
as if their data centers were running within one’s own computer.
• Many Internet commerce companies are now running their applications in the
“cloud”—such as in Amazon’s EC2, in Google’s Application Engine, or in
Microsoft’s Azure. Many companies and universities have also migrated their Internet applications (e.g., email and Web hosting) to the cloud. Cloud companies not
only provide applications scalable computing and storage environments, but also
provide the applications implicit access to their high-performance private networks.
1.8 Summary
In this chapter we’ve covered a tremendous amount of material! We’ve looked at the
various pieces of hardware and software that make up the Internet in particular and
computer networks in general. We started at the edge of the network, looking at end
systems and applications, and at the transport service provided to the applications
running on the end systems. We also looked at the link-layer technologies and physical media typically found in the access network. We then dove deeper inside the
network, into the network core, identifying packet switching and circuit switching
as the two basic approaches for transporting data through a telecommunication network, and we examined the strengths and weaknesses of each approach. We also
examined the structure of the global Internet, learning that the Internet is a network
of networks. We saw that the Internet’s hierarchical structure, consisting of higherand lower-tier ISPs, has allowed it to scale to include thousands of networks.
In the second part of this introductory chapter, we examined several topics central
to the field of computer networking. We first examined the causes of delay, throughput and packet loss in a packet-switched network. We developed simple quantitative
models for transmission, propagation, and queuing delays as well as for throughput;
we’ll make extensive use of these delay models in the homework problems throughout this book. Next we examined protocol layering and service models, key architectural principles in networking that we will also refer back to throughout this book. We
also surveyed some of the more prevalent security attacks in the Internet day. We finished our introduction to networking with a brief history of computer networking. The
first chapter in itself constitutes a mini-course in computer networking.
So, we have indeed covered a tremendous amount of ground in this first chapter!
If you’re a bit overwhelmed, don’t worry. In the following chapters we’ll revisit all of
these ideas, covering them in much more detail (that’s a promise, not a threat!). At this
point, we hope you leave this chapter with a still-developing intuition for the pieces
1.8
that make up a network, a still-developing command of the vocabulary of networking
(don’t be shy about referring back to this chapter), and an ever-growing desire to learn
more about networking. That’s the task ahead of us for the rest of this book.
Road-Mapping This Book
Before starting any trip, you should always glance at a road map in order to become
familiar with the major roads and junctures that lie ahead. For the trip we are about
to embark on, the ultimate destination is a deep understanding of the how, what, and
why of computer networks. Our road map is the sequence of chapters of this book:
1.
2.
3.
4.
5.
6.
7.
8.
9.
Computer Networks and the Internet
Application Layer
Transport Layer
Network Layer
Link Layer and Local Area Networks
Wireless and Mobile Networks
Multimedia Networking
Security in Computer Networks
Network Management
Chapters 2 through 5 are the four core chapters of this book. You should notice
that these chapters are organized around the top four layers of the five-layer Internet
protocol stack, one chapter for each layer. Further note that our journey will begin at
the top of the Internet protocol stack, namely, the application layer, and will work
its way downward. The rationale behind this top-down journey is that once we
understand the applications, we can understand the network services needed to support these applications. We can then, in turn, examine the various ways in which
such services might be implemented by a network architecture. Covering applications early thus provides motivation for the remainder of the text.
The second half of the book—Chapters 6 through 9—zooms in on four enormously important (and somewhat independent) topics in modern computer networking. In Chapter 6, we examine wireless and mobile networks, including wireless
LANs (including WiFi and Bluetooth), Cellular telephony networks (including
GSM, 3G, and 4G), and mobility (in both IP and GSM networks). In Chapter 7
(Multimedia Networking) we examine audio and video applications such as Internet
phone, video conferencing, and streaming of stored media. We also look at how a
packet-switched network can be designed to provide consistent quality of service to
audio and video applications. In Chapter 8 (Security in Computer Networks), we
first look at the underpinnings of encryption and network security, and then we
examine how the basic theory is being applied in a broad range of Internet contexts.
The last chapter (Network Management) examines the key issues in network management as well as the primary Internet protocols used for network management.
•
SUMMARY
67
68
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
Homework Problems and Questions
Chapter 1 Review Questions
SECTION 1.1
R1. What is the difference between a host and an end system? List several different
types of end systems. Is a Web server an end system?
R2. The word protocol is often used to describe diplomatic relations. How does
Wikipedia describe diplomatic protocol?
R3. Why are standards important for protocols?
SECTION 1.2
R4. List six access technologies. Classify each one as home access, enterprise
access, or wide-area wireless access.
R5. Is HFC transmission rate dedicated or shared among users? Are collisions
possible in a downstream HFC channel? Why or why not?
R6. List the available residential access technologies in your city. For each type
of access, provide the advertised downstream rate, upstream rate, and
monthly price.
R7. What is the transmission rate of Ethernet LANs?
R8. What are some of the physical media that Ethernet can run over?
R9. Dial-up modems, HFC, DSL and FTTH are all used for residential access.
For each of these access technologies, provide a range of transmission rates
and comment on whether the transmission rate is shared or dedicated.
R10. Describe the most popular wireless Internet access technologies today. Compare and contrast them.
SECTION 1.3
R11. Suppose there is exactly one packet switch between a sending host and a
receiving host. The transmission rates between the sending host and the
switch and between the switch and the receiving host are R1 and R2, respectively. Assuming that the switch uses store-and-forward packet switching,
what is the total end-to-end delay to send a packet of length L? (Ignore
queuing, propagation delay, and processing delay.)
R12. What advantage does a circuit-switched network have over a packet-switched
network? What advantages does TDM have over FDM in a circuit-switched
network?
R13. Suppose users share a 2 Mbps link. Also suppose each user transmits continuously at 1 Mbps when transmitting, but each user transmits only 20 percent of
the time. (See the discussion of statistical multiplexing in Section 1.3.)
HOMEWORK PROBLEMS AND QUESTIONS
a. When circuit switching is used, how many users can be supported?
b. For the remainder of this problem, suppose packet switching is used. Why
will there be essentially no queuing delay before the link if two or fewer
users transmit at the same time? Why will there be a queuing delay if
three users transmit at the same time?
c. Find the probability that a given user is transmitting.
d. Suppose now there are three users. Find the probability that at any given
time, all three users are transmitting simultaneously. Find the fraction of
time during which the queue grows.
R14. Why will two ISPs at the same level of the hierarchy often peer with each
other? How does an IXP earn money?
R15. Some content providers have created their own networks. Describe Google’s
network. What motivates content providers to create these networks?
SECTION 1.4
R16. Consider sending a packet from a source host to a destination host over a
fixed route. List the delay components in the end-to-end delay. Which of
these delays are constant and which are variable?
R17. Visit the Transmission Versus Propagation Delay applet at the companion
Web site. Among the rates, propagation delay, and packet sizes available,
find a combination for which the sender finishes transmitting before the first
bit of the packet reaches the receiver. Find another combination for which
the first bit of the packet reaches the receiver before the sender finishes
transmitting.
R18. How long does it take a packet of length 1,000 bytes to propagate over a link
of distance 2,500 km, propagation speed 2.5 · 108 m/s, and transmission rate
2 Mbps? More generally, how long does it take a packet of length L to propagate over a link of distance d, propagation speed s, and transmission rate R
bps? Does this delay depend on packet length? Does this delay depend on
transmission rate?
R19. Suppose Host A wants to send a large file to Host B. The path from Host A to
Host B has three links, of rates R1 = 500 kbps, R2 = 2 Mbps, and R3 = 1 Mbps.
a. Assuming no other traffic in the network, what is the throughput for the
file transfer?
b. Suppose the file is 4 million bytes. Dividing the file size by the throughput,
roughly how long will it take to transfer the file to Host B?
c. Repeat (a) and (b), but now with R2 reduced to 100 kbps.
R20. Suppose end system A wants to send a large file to end system B. At a very
high level, describe how end system A creates packets from the file. When
69
70
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
one of these packets arrives to a packet switch, what information in the
packet does the switch use to determine the link onto which the packet is
forwarded? Why is packet switching in the Internet analogous to driving from
one city to another and asking directions along the way?
R21. Visit the Queuing and Loss applet at the companion Web site. What is the
maximum emission rate and the minimum transmission rate? With those
rates, what is the traffic intensity? Run the applet with these rates and determine how long it takes for packet loss to occur. Then repeat the experiment a
second time and determine again how long it takes for packet loss to occur.
Are the values different? Why or why not?
SECTION 1.5
R22. List five tasks that a layer can perform. Is it possible that one (or more) of
these tasks could be performed by two (or more) layers?
R23. What are the five layers in the Internet protocol stack? What are the principal
responsibilities of each of these layers?
R24. What is an application-layer message? A transport-layer segment? A networklayer datagram? A link-layer frame?
R25. Which layers in the Internet protocol stack does a router process? Which
layers does a link-layer switch process? Which layers does a host process?
SECTION 1.6
R26. What is the difference between a virus and a worm?
R27. Describe how a botnet can be created, and how it can be used for a DDoS
attack.
R28. Suppose Alice and Bob are sending packets to each other over a computer
network. Suppose Trudy positions herself in the network so that she can
capture all the packets sent by Alice and send whatever she wants to Bob;
she can also capture all the packets sent by Bob and send whatever she
wants to Alice. List some of the malicious things Trudy can do from this
position.
Problems
P1. Design and describe an application-level protocol to be used between an
automatic teller machine and a bank’s centralized computer. Your protocol
should allow a user’s card and password to be verified, the account balance
(which is maintained at the centralized computer) to be queried, and an
account withdrawal to be made (that is, money disbursed to the user). Your
PROBLEMS
P2.
P3.
P4.
P5.
protocol entities should be able to handle the all-too-common case in which
there is not enough money in the account to cover the withdrawal. Specify
your protocol by listing the messages exchanged and the action taken by the
automatic teller machine or the bank’s centralized computer on transmission
and receipt of messages. Sketch the operation of your protocol for the case of a
simple withdrawal with no errors, using a diagram similar to that in Figure 1.2.
Explicitly state the assumptions made by your protocol about the underlying
end-to-end transport service.
Equation 1.1 gives a formula for the end-to-end delay of sending one packet
of length L over N links of transmission rate R. Generalize this formula for
sending P such packets back-to-back over the N links.
Consider an application that transmits data at a steady rate (for example, the
sender generates an N-bit unit of data every k time units, where k is small and
fixed). Also, when such an application starts, it will continue running for a
relatively long period of time. Answer the following questions, briefly justifying your answer:
a. Would a packet-switched network or a circuit-switched network be more
appropriate for this application? Why?
b. Suppose that a packet-switched network is used and the only traffic in
this network comes from such applications as described above. Furthermore, assume that the sum of the application data rates is less than the
capacities of each and every link. Is some form of congestion control
needed? Why?
Consider the circuit-switched network in Figure 1.13. Recall that there are 4
circuits on each link. Label the four switches A, B, C and D, going in the
clockwise direction.
a. What is the maximum number of simultaneous connections that can be in
progress at any one time in this network?
b. Suppose that all connections are between switches A and C. What is the
maximum number of simultaneous connections that can be in progress?
c. Suppose we want to make four connections between switches A and C,
and another four connections between switches B and D. Can we route
these calls through the four links to accommodate all eight connections?
Review the car-caravan analogy in Section 1.4. Assume a propagation speed
of 100 km/hour.
a. Suppose the caravan travels 150 km, beginning in front of one tollbooth,
passing through a second tollbooth, and finishing just after a third tollbooth. What is the end-to-end delay?
b. Repeat (a), now assuming that there are eight cars in the caravan instead
of ten.
71
72
CHAPTER 1
VideoNote
Exploring propagation
delay and transmission
delay
•
COMPUTER NETWORKS AND THE INTERNET
P6. This elementary problem begins to explore propagation delay and transmission delay, two central concepts in data networking. Consider two hosts, A
and B, connected by a single link of rate R bps. Suppose that the two hosts
are separated by m meters, and suppose the propagation speed along the link
is s meters/sec. Host A is to send a packet of size L bits to Host B.
a. Express the propagation delay, dprop, in terms of m and s.
b. Determine the transmission time of the packet, dtrans, in terms of L
and R.
c. Ignoring processing and queuing delays, obtain an expression for the endto-end delay.
d. Suppose Host A begins to transmit the packet at time t = 0. At time t = dtrans,
where is the last bit of the packet?
e. Suppose dprop is greater than dtrans. At time t = dtrans, where is the first bit of
the packet?
f. Suppose dprop is less than dtrans. At time t = dtrans, where is the first bit of
the packet?
g. Suppose s = 2.5 · 108, L = 120 bits, and R = 56 kbps. Find the distance m
so that dprop equals dtrans.
P7. In this problem, we consider sending real-time voice from Host A to Host B
over a packet-switched network (VoIP). Host A converts analog voice to a
digital 64 kbps bit stream on the fly. Host A then groups the bits into 56-byte
packets. There is one link between Hosts A and B; its transmission rate is 2
Mbps and its propagation delay is 10 msec. As soon as Host A gathers a
packet, it sends it to Host B. As soon as Host B receives an entire packet, it
converts the packet’s bits to an analog signal. How much time elapses from
the time a bit is created (from the original analog signal at Host A) until the
bit is decoded (as part of the analog signal at Host B)?
P8. Suppose users share a 3 Mbps link. Also suppose each user requires
150 kbps when transmitting, but each user transmits only 10 percent of the
time. (See the discussion of packet switching versus circuit switching in
Section 1.3.)
a. When circuit switching is used, how many users can be supported?
b. For the remainder of this problem, suppose packet switching is used. Find
the probability that a given user is transmitting.
c. Suppose there are 120 users. Find the probability that at any given time,
exactly n users are transmitting simultaneously. (Hint: Use the binomial
distribution.)
d. Find the probability that there are 21 or more users transmitting
simultaneously.
PROBLEMS
P9. Consider the discussion in Section 1.3 of packet switching versus circuit
switching in which an example is provided with a 1 Mbps link. Users are
generating data at a rate of 100 kbps when busy, but are busy generating
data only with probability p = 0.1. Suppose that the 1 Mbps link is replaced
by a 1 Gbps link.
a. What is N, the maximum number of users that can be supported
simultaneously under circuit switching?
b. Now consider packet switching and a user population of M users. Give a
formula (in terms of p, M, N) for the probability that more than N users are
sending data.
P10. Consider a packet of length L which begins at end system A and travels over
three links to a destination end system. These three links are connected by
two packet switches. Let di, si, and Ri denote the length, propagation speed,
and the transmission rate of link i, for i = 1, 2, 3. The packet switch delays
each packet by dproc. Assuming no queuing delays, in terms of di, si, Ri,
(i = 1,2,3), and L, what is the total end-to-end delay for the packet? Suppose
now the packet is 1,500 bytes, the propagation speed on all three links is 2.5 ·
108 m/s, the transmission rates of all three links are 2 Mbps, the packet switch
processing delay is 3 msec, the length of the first link is 5,000 km, the length
of the second link is 4,000 km, and the length of the last link is 1,000 km. For
these values, what is the end-to-end delay?
P11. In the above problem, suppose R1 = R2 = R3 = R and dproc = 0. Further suppose the packet switch does not store-and-forward packets but instead immediately transmits each bit it receives before waiting for the entire packet to
arrive. What is the end-to-end delay?
P12. A packet switch receives a packet and determines the outbound link to which
the packet should be forwarded. When the packet arrives, one other packet is
halfway done being transmitted on this outbound link and four other packets
are waiting to be transmitted. Packets are transmitted in order of arrival.
Suppose all packets are 1,500 bytes and the link rate is 2 Mbps. What is the
queuing delay for the packet? More generally, what is the queuing delay
when all packets have length L, the transmission rate is R, x bits of the
currently-being-transmitted packet have been transmitted, and n packets are
already in the queue?
P13. (a) Suppose N packets arrive simultaneously to a link at which no packets
are currently being transmitted or queued. Each packet is of length L
and the link has transmission rate R. What is the average queuing delay
for the N packets?
(b) Now suppose that N such packets arrive to the link every LN/R seconds.
What is the average queuing delay of a packet?
73
74
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
P14. Consider the queuing delay in a router buffer. Let I denote traffic intensity;
that is, I = La/R. Suppose that the queuing delay takes the form IL/R (1 – I)
for I < 1.
a. Provide a formula for the total delay, that is, the queuing delay plus the
transmission delay.
b. Plot the total delay as a function of L/R.
P15. Let a denote the rate of packets arriving at a link in packets/sec, and let μ
denote the link’s transmission rate in packets/sec. Based on the formula for
the total delay (i.e., the queuing delay plus the transmission delay) derived in
the previous problem, derive a formula for the total delay in terms of a and μ.
P16. Consider a router buffer preceding an outbound link. In this problem, you will
use Little’s formula, a famous formula from queuing theory. Let N denote the
average number of packets in the buffer plus the packet being transmitted. Let
a denote the rate of packets arriving at the link. Let d denote the average total
delay (i.e., the queuing delay plus the transmission delay) experienced by a
packet. Little’s formula is N = a · d. Suppose that on average, the buffer contains 10 packets, and the average packet queuing delay is 10 msec. The link’s
transmission rate is 100 packets/sec. Using Little’s formula, what is the average packet arrival rate, assuming there is no packet loss?
VideoNote
Using Traceroute to
discover network
paths and measure
network delay
P17. a. Generalize Equation 1.2 in Section 1.4.3 for heterogeneous processing
rates, transmission rates, and propagation delays.
b. Repeat (a), but now also suppose that there is an average queuing delay of
dqueue at each node.
P18. Perform a Traceroute between source and destination on the same continent
at three different hours of the day.
a. Find the average and standard deviation of the round-trip delays at each of
the three hours.
b. Find the number of routers in the path at each of the three hours. Did the
paths change during any of the hours?
c. Try to identify the number of ISP networks that the Traceroute packets pass
through from source to destination. Routers with similar names and/or similar
IP addresses should be considered as part of the same ISP. In your experiments,
do the largest delays occur at the peering interfaces between adjacent ISPs?
d. Repeat the above for a source and destination on different continents.
Compare the intra-continent and inter-continent results.
P19. (a) Visit the site www.traceroute.org and perform traceroutes from two different
cities in France to the same destination host in the United States. How many
links are the same in the two traceroutes? Is the transatlantic link the same?
PROBLEMS
P20.
P21.
P22.
P23.
P24.
(b) Repeat (a) but this time choose one city in France and another city in
Germany.
(c) Pick a city in the United States, and perform traceroutes to two hosts, each
in a different city in China. How many links are common in the two
traceroutes? Do the two traceroutes diverge before reaching China?
Consider the throughput example corresponding to Figure 1.20(b). Now
suppose that there are M client-server pairs rather than 10. Denote Rs, Rc, and
R for the rates of the server links, client links, and network link. Assume all
other links have abundant capacity and that there is no other traffic in the
network besides the traffic generated by the M client-server pairs. Derive a
general expression for throughput in terms of Rs, Rc, R, and M.
Consider Figure 1.19(b). Now suppose that there are M paths between the
server and the client. No two paths share any link. Path k (k = 1, . . ., M ) consists of N links with transmission rates R1k, R2k, . . ., RNk . If the server can only
use one path to send data to the client, what is the maximum throughput that
the server can achieve? If the server can use all M paths to send data, what is
the maximum throughput that the server can achieve?
Consider Figure 1.19(b). Suppose that each link between the server and the
client has a packet loss probability p, and the packet loss probabilities for
these links are independent. What is the probability that a packet (sent by the
server) is successfully received by the receiver? If a packet is lost in the path
from the server to the client, then the server will re-transmit the packet. On
average, how many times will the server re-transmit the packet in order for
the client to successfully receive the packet?
Consider Figure 1.19(a). Assume that we know the bottleneck link along the
path from the server to the client is the first link with rate Rs bits/sec. Suppose
we send a pair of packets back to back from the server to the client, and there
is no other traffic on this path. Assume each packet of size L bits, and both
links have the same propagation delay dprop.
a. What is the packet inter-arrival time at the destination? That is, how much
time elapses from when the last bit of the first packet arrives until the last
bit of the second packet arrives?
b. Now assume that the second link is the bottleneck link (i.e., Rc < Rs). Is it
possible that the second packet queues at the input queue of the second
link? Explain. Now suppose that the server sends the second packet T seconds after sending the first packet. How large must T be to ensure no
queuing before the second link? Explain.
Suppose you would like to urgently deliver 40 terabytes data from Boston to
Los Angeles. You have available a 100 Mbps dedicated link for data transfer.
Would you prefer to transmit the data via this link or instead use FedEx overnight delivery? Explain.
75
76
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
P25. Suppose two hosts, A and B, are separated by 20,000 kilometers and are
connected by a direct link of R = 2 Mbps. Suppose the propagation speed
over the link is 2.5 ⭈ 108 meters/sec.
a. Calculate the bandwidth-delay product, R ⭈ dprop.
b. Consider sending a file of 800,000 bits from Host A to Host B. Suppose
the file is sent continuously as one large message. What is the maximum
number of bits that will be in the link at any given time?
c. Provide an interpretation of the bandwidth-delay product.
d. What is the width (in meters) of a bit in the link? Is it longer than a football field?
e. Derive a general expression for the width of a bit in terms of the propagation speed s, the transmission rate R, and the length of the link m.
P26. Referring to problem P25, suppose we can modify R. For what value of R is
the width of a bit as long as the length of the link?
P27. Consider problem P25 but now with a link of R = 1 Gbps.
a. Calculate the bandwidth-delay product, R ⭈ dprop.
b. Consider sending a file of 800,000 bits from Host A to Host B. Suppose
the file is sent continuously as one big message. What is the maximum
number of bits that will be in the link at any given time?
c. What is the width (in meters) of a bit in the link?
P28. Refer again to problem P25.
a. How long does it take to send the file, assuming it is sent continuously?
b. Suppose now the file is broken up into 20 packets with each packet containing 40,000 bits. Suppose that each packet is acknowledged by the
receiver and the transmission time of an acknowledgment packet is
negligible. Finally, assume that the sender cannot send a packet until the
preceding one is acknowledged. How long does it take to send the file?
c. Compare the results from (a) and (b).
P29. Suppose there is a 10 Mbps microwave link between a geostationary satellite
and its base station on Earth. Every minute the satellite takes a digital photo and
sends it to the base station. Assume a propagation speed of 2.4 ⭈ 108 meters/sec.
a. What is the propagation delay of the link?
b. What is the bandwidth-delay product, R · dprop?
c. Let x denote the size of the photo. What is the minimum value of x for the
microwave link to be continuously transmitting?
P30. Consider the airline travel analogy in our discussion of layering in Section
1.5, and the addition of headers to protocol data units as they flow down
PROBLEMS
the protocol stack. Is there an equivalent notion of header information that
is added to passengers and baggage as they move down the airline protocol
stack?
P31. In modern packet-switched networks, including the Internet, the source host
segments long, application-layer messages (for example, an image or a music
file) into smaller packets and sends the packets into the network. The receiver
then reassembles the packets back into the original message. We refer to this
process as message segmentation. Figure 1.27 illustrates the end-to-end
transport of a message with and without message segmentation. Consider a
message that is 8 · 106 bits long that is to be sent from source to destination in
Figure 1.27. Suppose each link in the figure is 2 Mbps. Ignore propagation,
queuing, and processing delays.
a. Consider sending the message from source to destination without message
segmentation. How long does it take to move the message from the source
host to the first packet switch? Keeping in mind that each switch uses
store-and-forward packet switching, what is the total time to move the
message from source host to destination host?
b. Now suppose that the message is segmented into 800 packets, with each
packet being 10,000 bits long. How long does it take to move the first
packet from source host to the first switch? When the first packet is being
sent from the first switch to the second switch, the second packet is being
sent from the source host to the first switch. At what time will the second
packet be fully received at the first switch?
c. How long does it take to move the file from source host to destination host
when message segmentation is used? Compare this result with your
answer in part (a) and comment.
Message
a.
Source
Packet switch
Packet switch
Destination
Packet
b.
Source
Figure 1.27
Packet switch
Packet switch
Destination
End-to-end message transport: (a) without message
segmentation; (b) with message segmentation
77
78
CHAPTER 1
•
COMPUTER NETWORKS AND THE INTERNET
d. In addition to reducing delay, what are reasons to use message segmentation?
e. Discuss the drawbacks of message segmentation.
P32. Experiment with the Message Segmentation applet at the book’s Web site. Do
the delays in the applet correspond to the delays in the previous problem?
How do link propagation delays affect the overall end-to-end delay for packet
switching (with message segmentation) and for message switching?
P33. Consider sending a large file of F bits from Host A to Host B. There are three
links (and two switches) between A and B, and the links are uncongested (that
is, no queuing delays). Host A segments the file into segments of S bits each and
adds 80 bits of header to each segment, forming packets of L = 80 + S bits. Each
link has a transmission rate of R bps. Find the value of S that minimizes the
delay of moving the file from Host A to Host B. Disregard propagation delay.
P34. Skype offers a service that allows you to make a phone call from a PC to an
ordinary phone. This means that the voice call must pass through both the
Internet and through a telephone network. Discuss how this might be done.
Wireshark Lab
“Tell me and I forget. Show me and I remember. Involve me and I understand.”
Chinese proverb
One’s understanding of network protocols can often be greatly deepened by seeing
them in action and by playing around with them—observing the sequence of messages exchanged between two protocol entities, delving into the details of protocol
operation, causing protocols to perform certain actions, and observing these actions
and their consequences. This can be done in simulated scenarios or in a real network
environment such as the Internet. The Java applets at the textbook Web site take the
first approach. In the Wireshark labs, we’ll take the latter approach. You’ll run network applications in various scenarios using a computer on your desk, at home, or
in a lab. You’ll observe the network protocols in your computer, interacting and
exchanging messages with protocol entities executing elsewhere in the Internet.
Thus, you and your computer will be an integral part of these live labs. You’ll
observe—and you’ll learn—by doing.
The basic tool for observing the messages exchanged between executing
protocol entities is called a packet sniffer. As the name suggests, a packet sniffer
passively copies (sniffs) messages being sent from and received by your computer;
it also displays the contents of the various protocol fields of these captured messages. A screenshot of the Wireshark packet sniffer is shown in Figure 1.28. Wireshark is a free packet sniffer that runs on Windows, Linux/Unix, and Mac
WIRESHARK LAB
Command
menus
Listing of
captured
packets
Details of
selected
packet
header
Packet
contents in
hexadecimal
and ASCII
Figure 1.28
A Wireshark screen shot (Wireshark screenshot reprinted
by permission of the Wireshark Foundation.)
computers. Throughout the textbook, you will find Wireshark labs that allow you to
explore a number of the protocols studied in the chapter. In this first Wireshark lab,
you’ll obtain and install a copy of Wireshark, access a Web site, and capture and
examine the protocol messages being exchanged between your Web browser and the
Web server.
You can find full details about this first Wireshark lab (including instructions
about how to obtain and install Wireshark) at the Web site http://www.awl.com/
kurose-ross.
79
AN INTERVIEW WITH...
Leonard Kleinrock
Leonard Kleinrock is a professor of computer science at the University
of California, Los Angeles. In 1969, his computer at UCLA became
the first node of the Internet. His creation of packet-switching principles in 1961 became the technology behind the Internet. He
received his B.E.E. from the City College of New York (CCNY) and
his masters and PhD in electrical engineering from MIT.
What made you decide to specialize in networking/Internet technology?
As a PhD student at MIT in 1959, I looked around and found that most of my classmates
were doing research in the area of information theory and coding theory. At MIT, there was
the great researcher, Claude Shannon, who had launched these fields and had solved most of
the important problems already. The research problems that were left were hard and of lesser consequence. So I decided to launch out in a new area that no one else had yet conceived
of. Remember that at MIT I was surrounded by lots of computers, and it was clear to me
that soon these machines would need to communicate with each other. At the time, there
was no effective way for them to do so, so I decided to develop the technology that would
permit efficient and reliable data networks to be created.
What was your first job in the computer industry? What did it entail?
I went to the evening session at CCNY from 1951 to 1957 for my bachelor’s degree in
electrical engineering. During the day, I worked first as a technician and then as an engineer at a small, industrial electronics firm called Photobell. While there, I introduced
digital technology to their product line. Essentially, we were using photoelectric devices
to detect the presence of certain items (boxes, people, etc.) and the use of a circuit
known then as a bistable multivibrator was just the kind of technology we needed to
bring digital processing into this field of detection. These circuits happen to be the building blocks for computers, and have come to be known as flip-flops or switches in today’s
vernacular.
What was going through your mind when you sent the first host-to-host message (from
UCLA to the Stanford Research Institute)?
Frankly, we had no idea of the importance of that event. We had not prepared a special message of historic significance, as did so many inventors of the past (Samuel Morse with “What
hath God wrought.” or Alexander Graham Bell with “Watson, come here! I want you.” or
Neal Amstrong with “That’s one small step for a man, one giant leap for mankind.”) Those
guys were smart! They understood media and public relations. All we wanted to do was to
login to the SRI computer. So we typed the “L”, which was correctly received, we typed the
“o” which was received, and then we typed the “g” which caused the SRI host computer to
80
crash! So, it turned out that our message was the shortest and perhaps the most prophetic
message ever, namely “Lo!” as in “Lo and behold!”
Earlier that year, I was quoted in a UCLA press release saying that once the network was
up and running, it would be possible to gain access to computer utilities from our homes and
offices as easily as we gain access to electricity and telephone connectivity. So my vision at
that time was that the Internet would be ubiquitous, always on, always available, anyone with
any device could connect from any location, and it would be invisible. However, I never
anticipated that my 99-year-old mother would use the Internet—and indeed she did!
What is your vision for the future of networking?
The easy part of the vision is to predict the infrastructure itself. I anticipate that we see considerable deployment of nomadic computing, mobile devices, and smart spaces. Indeed, the
availability of lightweight, inexpensive, high-performance, portable computing, and communication devices (plus the ubiquity of the Internet) has enabled us to become nomads.
Nomadic computing refers to the technology that enables end users who travel from place to
place to gain access to Internet services in a transparent fashion, no matter where they travel
and no matter what device they carry or gain access to. The harder part of the vision is to
predict the applications and services, which have consistently surprised us in dramatic ways
(email, search technologies, the world-wide-web, blogs, social networks, user generation, and
sharing of music, photos, and videos, etc.). We are on the verge of a new class of surprising
and innovative mobile applications delivered to our hand-held devices.
The next step will enable us to move out from the netherworld of cyberspace to the
physical world of smart spaces. Our environments (desks, walls, vehicles, watches, belts, and
so on) will come alive with technology, through actuators, sensors, logic, processing, storage,
cameras, microphones, speakers, displays, and communication. This embedded technology
will allow our environment to provide the IP services we want. When I walk into a room, the
room will know I entered. I will be able to communicate with my environment naturally, as
in spoken English; my requests will generate replies that present Web pages to me from wall
displays, through my eyeglasses, as speech, holograms, and so forth.
Looking a bit further out, I see a networking future that includes the following additional key components. I see intelligent software agents deployed across the network
whose function it is to mine data, act on that data, observe trends, and carry out tasks
dynamically and adaptively. I see considerably more network traffic generated not so much
by humans, but by these embedded devices and these intelligent software agents. I see
large collections of self-organizing systems controlling this vast, fast network. I see huge
amounts of information flashing across this network instantaneously with this information
undergoing enormous processing and filtering. The Internet will essentially be a pervasive
global nervous system. I see all these things and more as we move headlong through the
twenty-first century.
81
What people have inspired you professionally?
By far, it was Claude Shannon from MIT, a brilliant researcher who had the ability to relate
his mathematical ideas to the physical world in highly intuitive ways. He was on my PhD
thesis committee.
Do you have any advice for students entering the networking/Internet field?
The Internet and all that it enables is a vast new frontier, full of amazing challenges. There
is room for great innovation. Don’t be constrained by today’s technology. Reach out and
imagine what could be and then make it happen.
82
CHAPTER
2
Application
Layer
Network applications are the raisons d’être of a computer network—if we couldn’t
conceive of any useful applications, there wouldn’t be any need for networking protocols that support these applications. Since the Internet’s inception, numerous useful and
entertaining applications have indeed been created. These applications have been the
driving force behind the Internet’s success, motivating people in homes, schools, governments, and businesses to make the Internet an integral part of their daily activities.
Internet applications include the classic text-based applications that became
popular in the 1970s and 1980s: text email, remote access to computers, file transfers, and newsgroups. They include the killer application of the mid-1990s, the
World Wide Web, encompassing Web surfing, search, and electronic commerce.
They include instant messaging and P2P file sharing, the two killer applications
introduced at the end of the millennium. Since 2000, we have seen an explosion of
popular voice and video applications, including: voice-over-IP (VoIP) and video
conferencing over IP such as Skype; user-generated video distribution such as
YouTube; and movies on demand such as Netflix. During this same period we have
also seen the immergence of highly engaging multi-player online games, including
Second Life and World of Warcraft. And most recently, we have seen the emergence
of a new generation of social networking applications, such as Facebook and Twitter,
which have created engaging human networks on top of the Internet’s network of
routers and communication links. Clearly, there has been no slowing down of new
83
84
CHAPTER 2
•
APPLICATION LAYER
and exciting Internet applications. Perhaps some of the readers of this text will create the next generation of killer Internet applications!
In this chapter we study the conceptual and implementation aspects of network
applications. We begin by defining key application-layer concepts, including network
services required by applications, clients and servers, processes, and transport-layer
interfaces. We examine several network applications in detail, including the Web,
e-mail, DNS, and peer-to-peer (P2P) file distribution (Chapter 8 focuses on multimedia applications, including streaming video and VoIP). We then cover network application development, over both TCP and UDP. In particular, we study the socket API
and walk through some simple client-server applications in Python. We also provide
several fun and interesting socket programming assignments at the end of the chapter.
The application layer is a particularly good place to start our study of protocols.
It’s familiar ground. We’re acquainted with many of the applications that rely on the
protocols we’ll study. It will give us a good feel for what protocols are all about and
will introduce us to many of the same issues that we’ll see again when we study transport, network, and link layer protocols.
2.1 Principles of Network Applications
Suppose you have an idea for a new network application. Perhaps this application
will be a great service to humanity, or will please your professor, or will bring you
great wealth, or will simply be fun to develop. Whatever the motivation may be, let’s
now examine how you transform the idea into a real-world network application.
At the core of network application development is writing programs that run on
different end systems and communicate with each other over the network. For
example, in the Web application there are two distinct programs that communicate
with each other: the browser program running in the user’s host (desktop, laptop,
tablet, smartphone, and so on); and the Web server program running in the Web
server host. As another example, in a P2P file-sharing system there is a program in
each host that participates in the file-sharing community. In this case, the programs
in the various hosts may be similar or identical.
Thus, when developing your new application, you need to write software that
will run on multiple end systems. This software could be written, for example, in C,
Java, or Python. Importantly, you do not need to write software that runs on networkcore devices, such as routers or link-layer switches. Even if you wanted to write
application software for these network-core devices, you wouldn’t be able to do so.
As we learned in Chapter 1, and as shown earlier in Figure 1.24, network-core
devices do not function at the application layer but instead function at lower layers—
specifically at the network layer and below. This basic design—namely, confining
application software to the end systems—as shown in Figure 2.1, has facilitated the
rapid development and deployment of a vast array of network applications.
2.1
•
PRINCIPLES OF NETWORK APPLICATIONS
Application
Transport
Network
Link
Physical
National or
Global ISP
Mobile Network
Local or
Regional ISP
Home Network
Application
Transport
Network
Application
Link
Transport
Physical
Network
Enterprise Network
Link
Physical
Figure 2.1
Communication for a network application takes place
between end systems at the application layer
85
86
CHAPTER 2
•
APPLICATION LAYER
2.1.1 Network Application Architectures
Before diving into software coding, you should have a broad architectural plan for
your application. Keep in mind that an application’s architecture is distinctly different from the network architecture (e.g., the five-layer Internet architecture discussed
in Chapter 1). From the application developer’s perspective, the network architecture is fixed and provides a specific set of services to applications. The application
architecture, on the other hand, is designed by the application developer and dictates how the application is structured over the various end systems. In choosing the
application architecture, an application developer will likely draw on one of the two
predominant architectural paradigms used in modern network applications: the
client-server architecture or the peer-to-peer (P2P) architecture
In a client-server architecture, there is an always-on host, called the server,
which services requests from many other hosts, called clients. A classic example is the
Web application for which an always-on Web server services requests from browsers
running on client hosts. When a Web server receives a request for an object from a
client host, it responds by sending the requested object to the client host. Note that
with the client-server architecture, clients do not directly communicate with each
other; for example, in the Web application, two browsers do not directly communicate. Another characteristic of the client-server architecture is that the server has a
fixed, well-known address, called an IP address (which we’ll discuss soon). Because
the server has a fixed, well-known address, and because the server is always on, a
client can always contact the server by sending a packet to the server’s IP address.
Some of the better-known applications with a client-server architecture include the
Web, FTP, Telnet, and e-mail. The client-server architecture is shown in Figure 2.2(a).
Often in a client-server application, a single-server host is incapable of keeping up
with all the requests from clients. For example, a popular social-networking site can
quickly become overwhelmed if it has only one server handling all of its requests. For
this reason, a data center, housing a large number of hosts, is often used to create a
powerful virtual server. The most popular Internet services—such as search engines
(e.g., Google and Bing), Internet commerce (e.g., Amazon and e-Bay), Web-based
email (e.g., Gmail and Yahoo Mail), social networking (e.g., Facebook and Twitter)—
employ one or more data centers. As discussed in Section 1.3.3, Google has 30 to 50
data centers distributed around the world, which collectively handle search, YouTube,
Gmail, and other services. A data center can have hundreds of thousands of servers,
which must be powered and maintained. Additionally, the service providers must pay
recurring interconnection and bandwidth costs for sending data from their data centers.
In a P2P architecture, there is minimal (or no) reliance on dedicated servers in
data centers. Instead the application exploits direct communication between pairs of
intermittently connected hosts, called peers. The peers are not owned by the service
provider, but are instead desktops and laptops controlled by users, with most of the
peers residing in homes, universities, and offices. Because the peers communicate
without passing through a dedicated server, the architecture is called peer-to-peer.
Many of today’s most popular and traffic-intensive applications are based on P2P
architectures. These applications include file sharing (e.g., BitTorrent), peer-assisted
2.1
a. Client-server architecture
Figure 2.2
•
PRINCIPLES OF NETWORK APPLICATIONS
b. Peer-to-peer architecture
(a) Client-server architecture; (b) P2P architecture
download acceleration (e.g., Xunlei), Internet Telephony (e.g., Skype), and IPTV (e.g.,
Kankan and PPstream). The P2P architecture is illustrated in Figure 2.2(b). We mention that some applications have hybrid architectures, combining both client-server
and P2P elements. For example, for many instant messaging applications, servers are
used to track the IP addresses of users, but user-to-user messages are sent directly
between user hosts (without passing through intermediate servers).
One of the most compelling features of P2P architectures is their self-scalability.
For example, in a P2P file-sharing application, although each peer generates
workload by requesting files, each peer also adds service capacity to the system
by distributing files to other peers. P2P architectures are also cost effective, since
they normally don’t require significant server infrastructure and server bandwidth
(in contrast with clients-server designs with datacenters). However, future P2P
applications face three major challenges:
1. ISP Friendly. Most residential ISPs (including DSL and cable ISPs) have been
dimensioned for “asymmetrical” bandwidth usage, that is, for much more
87
88
CHAPTER 2
•
APPLICATION LAYER
downstream than upstream traffic. But P2P video streaming and file distribution applications shift upstream traffic from servers to residential ISPs, thereby
putting significant stress on the ISPs. Future P2P applications need to be
designed so that they are friendly to ISPs [Xie 2008].
2. Security. Because of their highly distributed and open nature, P2P applications
can be a challenge to secure [Doucer 2002; Yu 2006; Liang 2006; Naoumov
2006; Dhungel 2008; LeBlond 2011].
3. Incentives. The success of future P2P applications also depends on convincing
users to volunteer bandwidth, storage, and computation resources to the applications, which is the challenge of incentive design [Feldman 2005; Piatek
2008; Aperjis 2008; Liu 2010].
2.1.2 Processes Communicating
Before building your network application, you also need a basic understanding of
how the programs, running in multiple end systems, communicate with each other.
In the jargon of operating systems, it is not actually programs but processes that
communicate. A process can be thought of as a program that is running within an
end system. When processes are running on the same end system, they can communicate with each other with interprocess communication, using rules that are
governed by the end system’s operating system. But in this book we are not particularly interested in how processes in the same host communicate, but instead in
how processes running on different hosts (with potentially different operating systems) communicate.
Processes on two different end systems communicate with each other by exchanging messages across the computer network. A sending process creates and sends messages into the network; a receiving process receives these messages and possibly
responds by sending messages back. Figure 2.1 illustrates that processes communicating with each other reside in the application layer of the five-layer protocol stack.
Client and Server Processes
A network application consists of pairs of processes that send messages to each
other over a network. For example, in the Web application a client browser
process exchanges messages with a Web server process. In a P2P file-sharing system, a file is transferred from a process in one peer to a process in another peer.
For each pair of communicating processes, we typically label one of the two
processes as the client and the other process as the server. With the Web, a
browser is a client process and a Web server is a server process. With P2P file
sharing, the peer that is downloading the file is labeled as the client, and the peer
that is uploading the file is labeled as the server.
You may have observed that in some applications, such as in P2P file sharing, a
process can be both a client and a server. Indeed, a process in a P2P file-sharing system can both upload and download files. Nevertheless, in the context of any given
2.1
•
PRINCIPLES OF NETWORK APPLICATIONS
communication session between a pair of processes, we can still label one process
as the client and the other process as the server. We define the client and server
processes as follows:
In the context of a communication session between a pair of processes, the
process that initiates the communication (that is, initially contacts the other
process at the beginning of the session) is labeled as the client. The process
that waits to be contacted to begin the session is the server.
In the Web, a browser process initializes contact with a Web server process;
hence the browser process is the client and the Web server process is the server. In
P2P file sharing, when Peer A asks Peer B to send a specific file, Peer A is the client
and Peer B is the server in the context of this specific communication session. When
there’s no confusion, we’ll sometimes also use the terminology “client side and
server side of an application.” At the end of this chapter, we’ll step through simple
code for both the client and server sides of network applications.
The Interface Between the Process and the Computer Network
As noted above, most applications consist of pairs of communicating processes,
with the two processes in each pair sending messages to each other. Any message
sent from one process to another must go through the underlying network. A process
sends messages into, and receives messages from, the network through a software
interface called a socket. Let’s consider an analogy to help us understand processes
and sockets. A process is analogous to a house and its socket is analogous to its door.
When a process wants to send a message to another process on another host, it
shoves the message out its door (socket). This sending process assumes that there is
a transportation infrastructure on the other side of its door that will transport the
message to the door of the destination process. Once the message arrives at the destination host, the message passes through the receiving process’s door (socket), and
the receiving process then acts on the message
Figure 2.3 illustrates socket communication between two processes that communicate over the Internet. (Figure 2.3 assumes that the underlying transport protocol used by the processes is the Internet’s TCP protocol.) As shown in this
figure, a socket is the interface between the application layer and the transport
layer within a host. It is also referred to as the Application Programming Interface (API) between the application and the network, since the socket is the programming interface with which network applications are built. The application
developer has control of everything on the application-layer side of the socket but
has little control of the transport-layer side of the socket. The only control that the
application developer has on the transport-layer side is (1) the choice of transport
protocol and (2) perhaps the ability to fix a few transport-layer parameters such as
maximum buffer and maximum segment sizes (to be covered in Chapter 3). Once
the application developer chooses a transport protocol (if a choice is available),
89
90
CHAPTER 2
Controlled
by application
developer
Controlled
by operating
system
Figure 2.3
•
APPLICATION LAYER
Host or
server
Host or
server
Process
Process
Socket
Socket
TCP with
buffers,
variables
TCP with
buffers,
variables
Internet
Controlled
by application
developer
Controlled
by operating
system
Application processes, sockets, and underlying transport protocol
the application is built using the transport-layer services provided by that protocol. We’ll explore sockets in some detail in Section 2.7.
Addressing Processes
In order to send postal mail to a particular destination, the destination needs to have
an address. Similarly, in order for a process running on one host to send packets to a
process running on another host, the receiving process needs to have an address.
To identify the receiving process, two pieces of information need to be specified:
(1) the address of the host and (2) an identifier that specifies the receiving process
in the destination host.
In the Internet, the host is identified by its IP address. We’ll discuss IP
addresses in great detail in Chapter 4. For now, all we need to know is that an IP
address is a 32-bit quantity that we can think of as uniquely identifying the host.
In addition to knowing the address of the host to which a message is destined, the
sending process must also identify the receiving process (more specifically, the
receiving socket) running in the host. This information is needed because in general a host could be running many network applications. A destination port number serves this purpose. Popular applications have been assigned specific port
numbers. For example, a Web server is identified by port number 80. A mail
server process (using the SMTP protocol) is identified by port number 25. A list
of well-known port numbers for all Internet standard protocols can be found at
http://www.iana.org. We’ll examine port numbers in detail in Chapter 3.
2.1
•
PRINCIPLES OF NETWORK APPLICATIONS
2.1.3 Transport Services Available to Applications
Recall that a socket is the interface between the application process and the
transport-layer protocol. The application at the sending side pushes messages
through the socket. At the other side of the socket, the transport-layer protocol
has the responsibility of getting the messages to the socket of the receiving
process.
Many networks, including the Internet, provide more than one transport-layer
protocol. When you develop an application, you must choose one of the available
transport-layer protocols. How do you make this choice? Most likely, you would
study the services provided by the available transport-layer protocols, and then pick
the protocol with the services that best match your application’s needs. The situation is similar to choosing either train or airplane transport for travel between two
cities. You have to choose one or the other, and each transportation mode offers different services. (For example, the train offers downtown pickup and drop-off,
whereas the plane offers shorter travel time.)
What are the services that a transport-layer protocol can offer to applications
invoking it? We can broadly classify the possible services along four dimensions:
reliable data transfer, throughput, timing, and security.
Reliable Data Transfer
As discussed in Chapter 1, packets can get lost within a computer network. For
example, a packet can overflow a buffer in a router, or can be discarded by a host
or router after having some of its bits corrupted. For many applications—such as
electronic mail, file transfer, remote host access, Web document transfers, and
financial applications—data loss can have devastating consequences (in the latter
case, for either the bank or the customer!). Thus, to support these applications,
something has to be done to guarantee that the data sent by one end of the application is delivered correctly and completely to the other end of the application. If
a protocol provides such a guaranteed data delivery service, it is said to provide
reliable data transfer. One important service that a transport-layer protocol can
potentially provide to an application is process-to-process reliable data transfer.
When a transport protocol provides this service, the sending process can just pass
its data into the socket and know with complete confidence that the data will
arrive without errors at the receiving process.
When a transport-layer protocol doesn’t provide reliable data transfer, some of
the data sent by the sending process may never arrive at the receiving process. This
may be acceptable for loss-tolerant applications, most notably multimedia applications such as conversational audio/video that can tolerate some amount of data loss.
In these multimedia applications, lost data might result in a small glitch in the
audio/video—not a crucial impairment.
91
92
CHAPTER 2
•
APPLICATION LAYER
Throughput
In Chapter 1 we introduced the concept of available throughput, which, in the context of a communication session between two processes along a network path, is
the rate at which the sending process can deliver bits to the receiving process.
Because other sessions will be sharing the bandwidth along the network path, and
because these other sessions will be coming and going, the available throughput
can fluctuate with time. These observations lead to another natural service that a
transport-layer protocol could provide, namely, guaranteed available throughput
at some specified rate. With such a service, the application could request a guaranteed throughput of r bits/sec, and the transport protocol would then ensure that
the available throughput is always at least r bits/sec. Such a guaranteed throughput service would appeal to many applications. For example, if an Internet telephony application encodes voice at 32 kbps, it needs to send data into the network
and have data delivered to the receiving application at this rate. If the transport
protocol cannot provide this throughput, the application would need to encode at
a lower rate (and receive enough throughput to sustain this lower coding rate) or
may have to give up, since receiving, say, half of the needed throughput is of little
or no use to this Internet telephony application. Applications that have throughput
requirements are said to be bandwidth-sensitive applications. Many current
multimedia applications are bandwidth sensitive, although some multimedia
applications may use adaptive coding techniques to encode digitized voice or
video at a rate that matches the currently available throughput.
While bandwidth-sensitive applications have specific throughput requirements, elastic applications can make use of as much, or as little, throughput
as happens to be available. Electronic mail, file transfer, and Web transfers are
all elastic applications. Of course, the more throughput, the better. There’s
an adage that says that one cannot be too rich, too thin, or have too much
throughput!
Timing
A transport-layer protocol can also provide timing guarantees. As with throughput
guarantees, timing guarantees can come in many shapes and forms. An example
guarantee might be that every bit that the sender pumps into the socket arrives at the
receiver’s socket no more than 100 msec later. Such a service would be appealing to
interactive real-time applications, such as Internet telephony, virtual environments,
teleconferencing, and multiplayer games, all of which require tight timing constraints on data delivery in order to be effective. (See Chapter 7, [Gauthier 1999;
Ramjee 1994].) Long delays in Internet telephony, for example, tend to result in
unnatural pauses in the conversation; in a multiplayer game or virtual interactive
environment, a long delay between taking an action and seeing the response from
the environment (for example, from another player at the end of an end-to-end connection) makes the application feel less realistic. For non-real-time applications,
2.1
•
PRINCIPLES OF NETWORK APPLICATIONS
lower delay is always preferable to higher delay, but no tight constraint is placed on
the end-to-end delays.
Security
Finally, a transport protocol can provide an application with one or more security
services. For example, in the sending host, a transport protocol can encrypt all data
transmitted by the sending process, and in the receiving host, the transport-layer
protocol can decrypt the data before delivering the data to the receiving process.
Such a service would provide confidentiality between the two processes, even if the
data is somehow observed between sending and receiving processes. A transport
protocol can also provide other security services in addition to confidentiality,
including data integrity and end-point authentication, topics that we’ll cover in
detail in Chapter 8.
2.1.4 Transport Services Provided by the Internet
Up until this point, we have been considering transport services that a computer
network could provide in general. Let’s now get more specific and examine the
type of transport services provided by the Internet. The Internet (and, more generally, TCP/IP networks) makes two transport protocols available to applications,
UDP and TCP. When you (as an application developer) create a new network
application for the Internet, one of the first decisions you have to make is
whether to use UDP or TCP. Each of these protocols offers a different set of services to the invoking applications. Figure 2.4 shows the service requirements for
some selected applications.
Application
Data Loss
Throughput
Time-Sensitive
File transfer/download
No loss
Elastic
No
E-mail
No loss
Elastic
No
Web documents
No loss
Elastic (few kbps)
No
Internet telephony/
Video conferencing
Loss-tolerant
Audio: few kbps–1Mbps
Video: 10 kbps–5 Mbps
Yes: 100s of msec
Streaming stored
audio/video
Loss-tolerant
Same as above
Yes: few seconds
Interactive games
Loss-tolerant
Few kbps–10 kbps
Yes: 100s of msec
Instant messaging
No loss
Elastic
Yes and no
Figure 2.4
Requirements of selected network applications
93
94
CHAPTER 2
•
APPLICATION LAYER
TCP Services
The TCP service model includes a connection-oriented service and a reliable data
transfer service. When an application invokes TCP as its transport protocol, the
application receives both of these services from TCP.
• Connection-oriented service. TCP has the client and server exchange transportlayer control information with each other before the application-level messages
begin to flow. This so-called handshaking procedure alerts the client and server,
allowing them to prepare for an onslaught of packets. After the handshaking phase,
a TCP connection is said to exist between the sockets of the two processes. The
connection is a full-duplex connection in that the two processes can send messages
to each other over the connection at the same time. When the application finishes
sending messages, it must tear down the connection. In Chapter 3 we’ll discuss
connection-oriented service in detail and examine how it is implemented.
FOCUS ON SECURITY
SECURING TCP
Neither TCP nor UDP provide any encryption—the data that the sending process passes into its socket is the same data that travels over the network to the destination
process. So, for example, if the sending process sends a password in cleartext (i.e.,
unencrypted) into its socket, the cleartext password will travel over all the links between
sender and receiver, potentially getting sniffed and discovered at any of the intervening
links. Because privacy and other security issues have become critical for many applications, the Internet community has developed an enhancement for TCP, called Secure
Sockets Layer (SSL). TCP-enhanced-with-SSL not only does everything that traditional
TCP does but also provides critical process-to-process security services, including
encryption, data integrity, and end-point authentication. We emphasize that SSL is not
a third Internet transport protocol, on the same level as TCP and UDP, but instead is an
enhancement of TCP, with the enhancements being implemented in the application
layer. In particular, if an application wants to use the services of SSL, it needs to
include SSL code (existing, highly optimized libraries and classes) in both the client and
server sides of the application. SSL has its own socket API that is similar to the traditional TCP socket API. When an application uses SSL, the sending process passes cleartext
data to the SSL socket; SSL in the sending host then encrypts the data and passes the
encrypted data to the TCP socket. The encrypted data travels over the Internet to the
TCP socket in the receiving process. The receiving socket passes the encrypted data to
SSL, which decrypts the data. Finally, SSL passes the cleartext data through its SSL
socket to the receiving process. We’ll cover SSL in some detail in Chapter 8.
2.1
•
PRINCIPLES OF NETWORK APPLICATIONS
• Reliable data transfer service. The communicating processes can rely on TCP
to deliver all data sent without error and in the proper order. When one side of
the application passes a stream of bytes into a socket, it can count on TCP to
deliver the same stream of bytes to the receiving socket, with no missing or
duplicate bytes.
TCP also includes a congestion-control mechanism, a service for the general
welfare of the Internet rather than for the direct benefit of the communicating
processes. The TCP congestion-control mechanism throttles a sending process (client
or server) when the network is congested between sender and receiver. As we will
see in Chapter 3, TCP congestion control also attempts to limit each TCP connection
to its fair share of network bandwidth.
UDP Services
UDP is a no-frills, lightweight transport protocol, providing minimal services. UDP
is connectionless, so there is no handshaking before the two processes start to
communicate. UDP provides an unreliable data transfer service—that is, when a process
sends a message into a UDP socket, UDP provides no guarantee that the message
will ever reach the receiving process. Furthermore, messages that do arrive at the
receiving process may arrive out of order.
UDP does not include a congestion-control mechanism, so the sending side of
UDP can pump data into the layer below (the network layer) at any rate it pleases.
(Note, however, that the actual end-to-end throughput may be less than this rate due
to the limited transmission capacity of intervening links or due to congestion).
Services Not Provided by Internet Transport Protocols
We have organized transport protocol services along four dimensions: reliable data
transfer, throughput, timing, and security. Which of these services are provided by
TCP and UDP? We have already noted that TCP provides reliable end-to-end data
transfer. And we also know that TCP can be easily enhanced at the application layer
with SSL to provide security services. But in our brief description of TCP and UDP,
conspicuously missing was any mention of throughput or timing guarantees—services not provided by today’s Internet transport protocols. Does this mean that timesensitive applications such as Internet telephony cannot run in today’s Internet? The
answer is clearly no—the Internet has been hosting time-sensitive applications for
many years. These applications often work fairly well because they have been
designed to cope, to the greatest extent possible, with this lack of guarantee. We’ll
investigate several of these design tricks in Chapter 7. Nevertheless, clever design
has its limitations when delay is excessive, or the end-to-end throughput is limited.
In summary, today’s Internet can often provide satisfactory service to time-sensitive
applications, but it cannot provide any timing or throughput guarantees.
95
96
CHAPTER 2
•
APPLICATION LAYER
Application
Application-Layer Protocol
Underlying Transport Protocol
Electronic mail
SMTP [RFC 5321]
TCP
Remote terminal access
Telnet [RFC 854]
TCP
Web
HTTP [RFC 2616]
TCP
File transfer
FTP [RFC 959]
TCP
Streaming multimedia
HTTP (e.g., YouTube)
TCP
Internet telephony
SIP [RFC 3261], RTP [RFC 3550], or proprietary
(e.g., Skype)
UDP or TCP
Figure 2.5
Popular Internet applications, their application-layer
protocols, and their underlying transport protocols
Figure 2.5 indicates the transport protocols used by some popular Internet
applications. We see that e-mail, remote terminal access, the Web, and file transfer all use TCP. These applications have chosen TCP primarily because TCP provides reliable data transfer, guaranteeing that all data will eventually get to its
destination. Because Internet telephony applications (such as Skype) can often
tolerate some loss but require a minimal rate to be effective, developers of Internet telephony applications usually prefer to run their applications over UDP,
thereby circumventing TCP’s congestion control mechanism and packet overheads. But because many firewalls are configured to block (most types of) UDP
traffic, Internet telephony applications often are designed to use TCP as a backup
if UDP communication fails.
2.1.5 Application-Layer Protocols
We have just learned that network processes communicate with each other by sending messages into sockets. But how are these messages structured? What are the
meanings of the various fields in the messages? When do the processes send the messages? These questions bring us into the realm of application-layer protocols. An
application-layer protocol defines how an application’s processes, running on different end systems, pass messages to each other. In particular, an application-layer
protocol defines:
• The types of messages exchanged, for example, request messages and response
messages
• The syntax of the various message types, such as the fields in the message and
how the fields are delineated
2.1
•
PRINCIPLES OF NETWORK APPLICATIONS
• The semantics of the fields, that is, the meaning of the information in the fields
• Rules for determining when and how a process sends messages and responds to
messages
Some application-layer protocols are specified in RFCs and are therefore in the public
domain. For example, the Web’s application-layer protocol, HTTP (the HyperText
Transfer Protocol [RFC 2616]), is available as an RFC. If a browser developer follows
the rules of the HTTP RFC, the browser will be able to retrieve Web pages from
any Web server that has also followed the rules of the HTTP RFC. Many other
application-layer protocols are proprietary and intentionally not available in the public
domain. For example, Skype uses proprietary application-layer protocols.
It is important to distinguish between network applications and application-layer
protocols. An application-layer protocol is only one piece of a network application
(albeit, a very important piece of the application from our point of view!). Let’s look at
a couple of examples. The Web is a client-server application that allows users to
obtain documents from Web servers on demand. The Web application consists of
many components, including a standard for document formats (that is, HTML), Web
browsers (for example, Firefox and Microsoft Internet Explorer), Web servers (for
example, Apache and Microsoft servers), and an application-layer protocol. The
Web’s application-layer protocol, HTTP, defines the format and sequence of messages
exchanged between browser and Web server. Thus, HTTP is only one piece (albeit, an
important piece) of the Web application. As another example, an Internet e-mail application also has many components, including mail servers that house user mailboxes;
mail clients (such as Microsoft Outlook) that allow users to read and create messages; a
standard for defining the structure of an e-mail message; and application-layer protocols that define how messages are passed between servers, how messages are passed
between servers and mail clients, and how the contents of message headers are to be
interpreted. The principal application-layer protocol for electronic mail is SMTP
(Simple Mail Transfer Protocol) [RFC 5321]. Thus, e-mail’s principal application-layer
protocol, SMTP, is only one piece (albeit, an important piece) of the e-mail application.
2.1.6 Network Applications Covered in This Book
New public domain and proprietary Internet applications are being developed every
day. Rather than covering a large number of Internet applications in an encyclopedic manner, we have chosen to focus on a small number of applications that are both
pervasive and important. In this chapter we discuss five important applications: the
Web, file transfer, electronic mail, directory service, and P2P applications. We first
discuss the Web, not only because it is an enormously popular application, but also
because its application-layer protocol, HTTP, is straightforward and easy to understand. After covering the Web, we briefly examine FTP, because it provides a nice
contrast to HTTP. We then discuss electronic mail, the Internet’s first killer application. E-mail is more complex than the Web in the sense that it makes use of not one
97
98
CHAPTER 2
•
APPLICATION LAYER
but several application-layer protocols. After e-mail, we cover DNS, which provides
a directory service for the Internet. Most users do not interact with DNS directly;
instead, users invoke DNS indirectly through other applications (including the Web,
file transfer, and electronic mail). DNS illustrates nicely how a piece of core network functionality (network-name to network-address translation) can be implemented at the application layer in the Internet. Finally, we discuss in this chapter
several P2P applications, focusing on file sharing applications, and distributed
lookup services. In Chapter 7, we’ll cover multimedia applications, including
streaming video and voice-over-IP.
2.2 The Web and HTTP
Until the early 1990s the Internet was used primarily by researchers, academics, and
university students to log in to remote hosts, to transfer files from local hosts to remote
hosts and vice versa, to receive and send news, and to receive and send electronic
mail. Although these applications were (and continue to be) extremely useful, the
Internet was essentially unknown outside of the academic and research communities.
Then, in the early 1990s, a major new application arrived on the scene—the World
Wide Web [Berners-Lee 1994]. The Web was the first Internet application that caught
the general public’s eye. It dramatically changed, and continues to change, how people interact inside and outside their work environments. It elevated the Internet from
just one of many data networks to essentially the one and only data network.
Perhaps what appeals the most to users is that the Web operates on demand.
Users receive what they want, when they want it. This is unlike traditional broadcast radio and television, which force users to tune in when the content provider
makes the content available. In addition to being available on demand, the Web has
many other wonderful features that people love and cherish. It is enormously easy
for any individual to make information available over the Web—everyone can
become a publisher at extremely low cost. Hyperlinks and search engines help us
navigate through an ocean of Web sites. Graphics stimulate our senses. Forms,
JavaScript, Java applets, and many other devices enable us to interact with pages
and sites. And the Web serves as a platform for many killer applications emerging
after 2003, including YouTube, Gmail, and Facebook.
2.2.1 Overview of HTTP
The HyperText Transfer Protocol (HTTP), the Web’s application-layer protocol,
is at the heart of the Web. It is defined in [RFC 1945] and [RFC 2616]. HTTP is
implemented in two programs: a client program and a server program. The client
program and server program, executing on different end systems, talk to each other
by exchanging HTTP messages. HTTP defines the structure of these messages and
how the client and server exchange the messages. Before explaining HTTP in detail,
we should review some Web terminology.
2.2
•
THE WEB AND HTTP
A Web page (also called a document) consists of objects. An object is simply a
file—such as an HTML file, a JPEG image, a Java applet, or a video clip—that is
addressable by a single URL. Most Web pages consist of a base HTML file and
several referenced objects. For example, if a Web page contains HTML text and five
JPEG images, then the Web page has six objects: the base HTML file plus the five
images. The base HTML file references the other objects in the page with the
objects’ URLs. Each URL has two components: the hostname of the server that
houses the object and the object’s path name. For example, the URL
http://www.someSchool.edu/someDepartment/picture.gif
has www.someSchool.edu for a hostname and /someDepartment/
picture.gif for a path name. Because Web browsers (such as Internet Explorer
and Firefox) implement the client side of HTTP, in the context of the Web, we will use
the words browser and client interchangeably. Web servers, which implement the
server side of HTTP, house Web objects, each addressable by a URL. Popular Web
servers include Apache and Microsoft Internet Information Server.
HTTP defines how Web clients request Web pages from Web servers and how
servers transfer Web pages to clients. We discuss the interaction between client and
server in detail later, but the general idea is illustrated in Figure 2.6. When a user
requests a Web page (for example, clicks on a hyperlink), the browser sends HTTP
request messages for the objects in the page to the server. The server receives the
requests and responds with HTTP response messages that contain the objects.
HTTP uses TCP as its underlying transport protocol (rather than running on top of
UDP). The HTTP client first initiates a TCP connection with the server. Once the connection is established, the browser and the server processes access TCP through their
socket interfaces. As described in Section 2.1, on the client side the socket interface is
the door between the client process and the TCP connection; on the server side it is the
Server running
Apache Web server
HT
TP
q
re
H
s
ue
P
TT
HT
t
re
o
sp
e
ns
HT
TP
TP
re
re
qu
PC running
Internet Explorer
Figure 2.6
HTTP request-response behavior
sp
es
on
se
t
Linux running
Firefox
99
100
CHAPTER 2
•
APPLICATION LAYER
door between the server process and the TCP connection. The client sends HTTP
request messages into its socket interface and receives HTTP response messages from
its socket interface. Similarly, the HTTP server receives request messages from its
socket interface and sends response messages into its socket interface. Once the client
sends a message into its socket interface, the message is out of the client’s hands and is
“in the hands” of TCP. Recall from Section 2.1 that TCP provides a reliable data transfer service to HTTP. This implies that each HTTP request message sent by a client
process eventually arrives intact at the server; similarly, each HTTP response message
sent by the server process eventually arrives intact at the client. Here we see one of the
great advantages of a layered architecture—HTTP need not worry about lost data or the
details of how TCP recovers from loss or reordering of data within the network. That is
the job of TCP and the protocols in the lower layers of the protocol stack.
It is important to note that the server sends requested files to clients without storing any state information about the client. If a particular client asks for the same object
twice in a period of a few seconds, the server does not respond by saying that it just
served the object to the client; instead, the server resends the object, as it has completely forgotten what it did earlier. Because an HTTP server maintains no information about the clients, HTTP is said to be a stateless protocol. We also remark that the
Web uses the client-server application architecture, as described in Section 2.1. A Web
server is always on, with a fixed IP address, and it services requests from potentially
millions of different browsers.
2.2.2 Non-Persistent and Persistent Connections
In many Internet applications, the client and server communicate for an extended
period of time, with the client making a series of requests and the server responding to
each of the requests. Depending on the application and on how the application is being
used, the series of requests may be made back-to-back, periodically at regular intervals,
or intermittently. When this client-server interaction is taking place over TCP, the application developer needs to make an important decision––should each request/response
pair be sent over a separate TCP connection, or should all of the requests and their corresponding responses be sent over the same TCP connection? In the former approach,
the application is said to use non-persistent connections; and in the latter approach,
persistent connections. To gain a deep understanding of this design issue, let’s examine the advantages and disadvantages of persistent connections in the context of a specific application, namely, HTTP, which can use both non-persistent connections and
persistent connections. Although HTTP uses persistent connections in its default mode,
HTTP clients and servers can be configured to use non-persistent connections instead.
HTTP with Non-Persistent Connections
Let’s walk through the steps of transferring a Web page from server to client for the
case of non-persistent connections. Let’s suppose the page consists of a base HTML
2.2
•
THE WEB AND HTTP
file and 10 JPEG images, and that all 11 of these objects reside on the same server.
Further suppose the URL for the base HTML file is
http://www.someSchool.edu/someDepartment/home.index
Here is what happens:
1. The HTTP client process initiates a TCP connection to the server
www.someSchool.edu on port number 80, which is the default port number for HTTP. Associated with the TCP connection, there will be a socket at the
client and a socket at the server.
2. The HTTP client sends an HTTP request message to the server via its socket. The
request message includes the path name /someDepartment/home.index.
(We will discuss HTTP messages in some detail below.)
3. The HTTP server process receives the request message via its socket, retrieves
the object /someDepartment/home.index from its storage (RAM or
disk), encapsulates the object in an HTTP response message, and sends the
response message to the client via its socket.
4. The HTTP server process tells TCP to close the TCP connection. (But TCP
doesn’t actually terminate the connection until it knows for sure that the client
has received the response message intact.)
5. The HTTP client receives the response message. The TCP connection terminates. The message indicates that the encapsulated object is an HTML file. The
client extracts the file from the response message, examines the HTML file,
and finds references to the 10 JPEG objects.
6. The first four steps are then repeated for each of the referenced JPEG objects.
As the browser receives the Web page, it displays the page to the user. Two different browsers may interpret (that is, display to the user) a Web page in somewhat different ways. HTTP has nothing to do with how a Web page is interpreted by a client. The
HTTP specifications ([RFC 1945] and [RFC 2616]) define only the communication
protocol between the client HTTP program and the server HTTP program.
The steps above illustrate the use of non-persistent connections, where each TCP
connection is closed after the server sends the object—the connection does not persist
for other objects. Note that each TCP connection transports exactly one request message and one response message. Thus, in this example, when a user requests the Web
page, 11 TCP connections are generated.
In the steps described above, we were intentionally vague about whether the
client obtains the 10 JPEGs over 10 serial TCP connections, or whether some of the
JPEGs are obtained over parallel TCP connections. Indeed, users can configure
modern browsers to control the degree of parallelism. In their default modes, most
browsers open 5 to 10 parallel TCP connections, and each of these connections handles one request-response transaction. If the user prefers, the maximum number of
101
102
CHAPTER 2
•
APPLICATION LAYER
parallel connections can be set to one, in which case the 10 connections are established serially. As we’ll see in the next chapter, the use of parallel connections shortens the response time.
Before continuing, let’s do a back-of-the-envelope calculation to estimate the
amount of time that elapses from when a client requests the base HTML file until
the entire file is received by the client. To this end, we define the round-trip time
(RTT), which is the time it takes for a small packet to travel from client to server
and then back to the client. The RTT includes packet-propagation delays, packetqueuing delays in intermediate routers and switches, and packet-processing
delays. (These delays were discussed in Section 1.4.) Now consider what happens
when a user clicks on a hyperlink. As shown in Figure 2.7, this causes the browser
to initiate a TCP connection between the browser and the Web server; this
involves a “three-way handshake”—the client sends a small TCP segment to the
server, the server acknowledges and responds with a small TCP segment, and,
finally, the client acknowledges back to the server. The first two parts of the threeway handshake take one RTT. After completing the first two parts of the handshake, the client sends the HTTP request message combined with the third part of
Initiate TCP
connection
RTT
Request file
RTT
Time to transmit file
Entire file received
Time
at client
Figure 2.7
Time
at server
Back-of-the-envelope calculation for the time needed to
request and receive an HTML file
2.2
•
THE WEB AND HTTP
the three-way handshake (the acknowledgment) into the TCP connection. Once
the request message arrives at the server, the server sends the HTML file into the
TCP connection. This HTTP request/response eats up another RTT. Thus, roughly,
the total response time is two RTTs plus the transmission time at the server of the
HTML file.
HTTP with Persistent Connections
Non-persistent connections have some shortcomings. First, a brand-new connection must be established and maintained for each requested object. For each of
these connections, TCP buffers must be allocated and TCP variables must be kept
in both the client and server. This can place a significant burden on the Web server,
which may be serving requests from hundreds of different clients simultaneously.
Second, as we just described, each object suffers a delivery delay of two RTTs—
one RTT to establish the TCP connection and one RTT to request and receive an
object.
With persistent connections, the server leaves the TCP connection open after
sending a response. Subsequent requests and responses between the same client and
server can be sent over the same connection. In particular, an entire Web page (in
the example above, the base HTML file and the 10 images) can be sent over a single
persistent TCP connection. Moreover, multiple Web pages residing on the same
server can be sent from the server to the same client over a single persistent TCP
connection. These requests for objects can be made back-to-back, without waiting
for replies to pending requests (pipelining). Typically, the HTTP server closes a connection when it isn’t used for a certain time (a configurable timeout interval). When
the server receives the back-to-back requests, it sends the objects back-to-back. The
default mode of HTTP uses persistent connections with pipelining. We’ll quantitatively compare the performance of non-persistent and persistent connections in the
homework problems of Chapters 2 and 3. You are also encouraged to see [Heidemann 1997; Nielsen 1997].
2.2.3 HTTP Message Format
The HTTP specifications [RFC 1945; RFC 2616] include the definitions of the
HTTP message formats. There are two types of HTTP messages, request messages
and response messages, both of which are discussed below.
HTTP Request Message
Below we provide a typical HTTP request message:
GET /somedir/page.html HTTP/1.1
Host: www.someschool.edu
103
104
CHAPTER 2
•
APPLICATION LAYER
Connection: close
User-agent: Mozilla/5.0
Accept-language: fr
We can learn a lot by taking a close look at this simple request message. First of
all, we see that the message is written in ordinary ASCII text, so that your ordinary
computer-literate human being can read it. Second, we see that the message consists
of five lines, each followed by a carriage return and a line feed. The last line is followed by an additional carriage return and line feed. Although this particular request
message has five lines, a request message can have many more lines or as few as
one line. The first line of an HTTP request message is called the request line; the
subsequent lines are called the header lines. The request line has three fields: the
method field, the URL field, and the HTTP version field. The method field can take
on several different values, including GET, POST, HEAD, PUT, and DELETE.
The great majority of HTTP request messages use the GET method. The GET
method is used when the browser requests an object, with the requested object identified in the URL field. In this example, the browser is requesting the object
/somedir/page.html. The version is self-explanatory; in this example, the
browser implements version HTTP/1.1.
Now let’s look at the header lines in the example. The header line Host:
www.someschool.edu specifies the host on which the object resides. You might
think that this header line is unnecessary, as there is already a TCP connection in
place to the host. But, as we’ll see in Section 2.2.5, the information provided by the
host header line is required by Web proxy caches. By including the Connection:
close header line, the browser is telling the server that it doesn’t want to bother
with persistent connections; it wants the server to close the connection after sending
the requested object. The User-agent: header line specifies the user agent, that
is, the browser type that is making the request to the server. Here the user agent is
Mozilla/5.0, a Firefox browser. This header line is useful because the server can
actually send different versions of the same object to different types of user agents.
(Each of the versions is addressed by the same URL.) Finally, the Acceptlanguage: header indicates that the user prefers to receive a French version of
the object, if such an object exists on the server; otherwise, the server should send
its default version. The Accept-language: header is just one of many content
negotiation headers available in HTTP.
Having looked at an example, let’s now look at the general format of a request
message, as shown in Figure 2.8. We see that the general format closely follows our
earlier example. You may have noticed, however, that after the header lines (and the
additional carriage return and line feed) there is an “entity body.” The entity body is
empty with the GET method, but is used with the POST method. An HTTP client
often uses the POST method when the user fills out a form—for example, when a
user provides search words to a search engine. With a POST message, the user is still
requesting a Web page from the server, but the specific contents of the Web page
2.2
Request line
method
sp
URL
sp
Version
header field name: sp
value
cr lf
header field name: sp
value
cr lf
•
THE WEB AND HTTP
cr lf
Header lines
cr lf
Blank line
Entity body
Figure 2.8
General format of an HTTP request message
depend on what the user entered into the form fields. If the value of the method field
is POST, then the entity body contains what the user entered into the form fields.
We would be remiss if we didn’t mention that a request generated with a form
does not necessarily use the POST method. Instead, HTML forms often use the GET
method and include the inputted data (in the form fields) in the requested URL. For
example, if a form uses the GET method, has two fields, and the inputs to the two
fields are monkeys and bananas, then the URL will have the structure
www.somesite.com/animalsearch?monkeys&bananas. In your day-today Web surfing, you have probably noticed extended URLs of this sort.
The HEAD method is similar to the GET method. When a server receives a
request with the HEAD method, it responds with an HTTP message but it leaves out
the requested object. Application developers often use the HEAD method for debugging. The PUT method is often used in conjunction with Web publishing tools. It
allows a user to upload an object to a specific path (directory) on a specific Web
server. The PUT method is also used by applications that need to upload objects to
Web servers. The DELETE method allows a user, or an application, to delete an
object on a Web server.
HTTP Response Message
Below we provide a typical HTTP response message. This response message could
be the response to the example request message just discussed.
HTTP/1.1 200 OK
Connection: close
105
106
CHAPTER 2
•
APPLICATION LAYER
Date: Tue, 09 Aug 2011 15:44:04 GMT
Server: Apache/2.2.3 (CentOS)
Last-Modified: Tue, 09 Aug 2011 15:11:03 GMT
Content-Length: 6821
Content-Type: text/html
(data data data data data ...)
Let’s take a careful look at this response message. It has three sections: an initial status line, six header lines, and then the entity body. The entity body is the
meat of the message—it contains the requested object itself (represented by data
data data data data ...). The status line has three fields: the protocol version field, a status code, and a corresponding status message. In this example, the
status line indicates that the server is using HTTP/1.1 and that everything is OK
(that is, the server has found, and is sending, the requested object).
Now let’s look at the header lines. The server uses the Connection: close
header line to tell the client that it is going to close the TCP connection after sending
the message. The Date: header line indicates the time and date when the HTTP
response was created and sent by the server. Note that this is not the time when the
object was created or last modified; it is the time when the server retrieves the
object from its file system, inserts the object into the response message, and sends
the response message. The Server: header line indicates that the message was generated by an Apache Web server; it is analogous to the User-agent: header line
in the HTTP request message. The Last-Modified: header line indicates the
time and date when the object was created or last modified. The Last-Modified:
header, which we will soon cover in more detail, is critical for object caching, both in
the local client and in network cache servers (also known as proxy servers). The
Content-Length: header line indicates the number of bytes in the object being
sent. The Content-Type: header line indicates that the object in the entity body is
HTML text. (The object type is officially indicated by the Content-Type: header
and not by the file extension.)
Having looked at an example, let’s now examine the general format of a
response message, which is shown in Figure 2.9. This general format of the response
message matches the previous example of a response message. Let’s say a few additional words about status codes and their phrases. The status code and associated
phrase indicate the result of the request. Some common status codes and associated
phrases include:
• 200 OK: Request succeeded and the information is returned in the response.
• 301 Moved Permanently: Requested object has been permanently moved;
the new URL is specified in Location: header of the response message. The
client software will automatically retrieve the new URL.
2.2
Status line
version
sp status code
sp
phrase
header field name: sp
value
cr lf
header field name: sp
value
cr lf
•
THE WEB AND HTTP
107
cr lf
Header lines
cr lf
Blank line
Entity body
Figure 2.9
General format of an HTTP response message
• 400 Bad Request: This is a generic error code indicating that the request
could not be understood by the server.
• 404 Not Found: The requested document does not exist on this server.
• 505 HTTP Version Not Supported: The requested HTTP protocol
version is not supported by the server.
How would you like to see a real HTTP response message? This is highly recommended and very easy to do! First Telnet into your favorite Web server. Then
type in a one-line request message for some object that is housed on the server. For
example, if you have access to a command prompt, type:
telnet cis.poly.edu 80
GET /~ross/ HTTP/1.1
Host: cis.poly.edu
(Press the carriage return twice after typing the last line.) This opens a TCP connection to port 80 of the host cis.poly.edu and then sends the HTTP request message. You should see a response message that includes the base HTML file of
Professor Ross’s homepage. If you’d rather just see the HTTP message lines and not
receive the object itself, replace GET with HEAD. Finally, replace /~ross/ with
/~banana/ and see what kind of response message you get.
In this section we discussed a number of header lines that can be used within
HTTP request and response messages. The HTTP specification defines many, many
VideoNote
Using Wireshark to
investigate the
HTTP protocol
108
CHAPTER 2
•
APPLICATION LAYER
more header lines that can be inserted by browsers, Web servers, and network cache
servers. We have covered only a small number of the totality of header lines. We’ll
cover a few more below and another small number when we discuss network Web
caching in Section 2.2.5. A highly readable and comprehensive discussion of the HTTP
protocol, including its headers and status codes, is given in [Krishnamurthy 2001].
How does a browser decide which header lines to include in a request message? How does a Web server decide which header lines to include in a response
message? A browser will generate header lines as a function of the browser type
and version (for example, an HTTP/1.0 browser will not generate any 1.1 header
lines), the user configuration of the browser (for example, preferred language), and
whether the browser currently has a cached, but possibly out-of-date, version of the
object. Web servers behave similarly: There are different products, versions, and
configurations, all of which influence which header lines are included in response
messages.
2.2.4 User-Server Interaction: Cookies
We mentioned above that an HTTP server is stateless. This simplifies server design
and has permitted engineers to develop high-performance Web servers that can handle thousands of simultaneous TCP connections. However, it is often desirable for a
Web site to identify users, either because the server wishes to restrict user access or
because it wants to serve content as a function of the user identity. For these purposes, HTTP uses cookies. Cookies, defined in [RFC 6265], allow sites to keep
track of users. Most major commercial Web sites use cookies today.
As shown in Figure 2.10, cookie technology has four components: (1) a cookie
header line in the HTTP response message; (2) a cookie header line in the HTTP
request message; (3) a cookie file kept on the user’s end system and managed by the
user’s browser; and (4) a back-end database at the Web site. Using Figure 2.10, let’s
walk through an example of how cookies work. Suppose Susan, who always
accesses the Web using Internet Explorer from her home PC, contacts Amazon.com
for the first time. Let us suppose that in the past she has already visited the eBay site.
When the request comes into the Amazon Web server, the server creates a unique
identification number and creates an entry in its back-end database that is indexed
by the identification number. The Amazon Web server then responds to Susan’s
browser, including in the HTTP response a Set-cookie: header, which contains
the identification number. For example, the header line might be:
Set-cookie: 1678
When Susan’s browser receives the HTTP response message, it sees the Setcookie: header. The browser then appends a line to the special cookie file that it
manages. This line includes the hostname of the server and the identification number in the Set-cookie: header. Note that the cookie file already has an entry for
2.2
Client host
l ht
tp
ebay: 8734
requ
est
Server creates
ID 1678 for user
resp
http 1678
l
a
u
us
ie:
cook
Setusua
l
cook http req
ie:
1678 uest msg
One week later
tp
l ht
re
entry in backend
database
Cookie-specific
action
sg
se m
spon
usua
l
cook http req
ie:
1678 uest msg
amazon: 1678
ebay: 8734
usua
l
resp
http
onse
Cookie-specific
action
Time
Key:
Cookie file
Keeping user state with cookies
eBay, since Susan has visited that site in the past. As Susan continues to browse the
Amazon site, each time she requests a Web page, her browser consults her cookie
file, extracts her identification number for this site, and puts a cookie header line
that includes the identification number in the HTTP request. Specifically, each of
her HTTP requests to the Amazon server includes the header line:
Cookie: 1678
access
access
msg
Time
Figure 2.10
109
msg
onse
usua
THE WEB AND HTTP
Server host
usua
amazon: 1678
ebay: 8734
•
110
CHAPTER 2
•
APPLICATION LAYER
In this manner, the Amazon server is able to track Susan’s activity at the Amazon
site. Although the Amazon Web site does not necessarily know Susan’s name, it
knows exactly which pages user 1678 visited, in which order, and at what times!
Amazon uses cookies to provide its shopping cart service—Amazon can maintain a
list of all of Susan’s intended purchases, so that she can pay for them collectively at
the end of the session.
If Susan returns to Amazon’s site, say, one week later, her browser will continue
to put the header line Cookie: 1678 in the request messages. Amazon also recommends products to Susan based on Web pages she has visited at Amazon in the
past. If Susan also registers herself with Amazon—providing full name, e-mail
address, postal address, and credit card information—Amazon can then include this
information in its database, thereby associating Susan’s name with her identification
number (and all of the pages she has visited at the site in the past!). This is how
Amazon and other e-commerce sites provide “one-click shopping”—when Susan
chooses to purchase an item during a subsequent visit, she doesn’t need to re-enter
her name, credit card number, or address.
From this discussion we see that cookies can be used to identify a user. The first
time a user visits a site, the user can provide a user identification (possibly his or her
name). During the subsequent sessions, the browser passes a cookie header to the
server, thereby identifying the user to the server. Cookies can thus be used to create
a user session layer on top of stateless HTTP. For example, when a user logs in to a
Web-based e-mail application (such as Hotmail), the browser sends cookie information to the server, permitting the server to identify the user throughout the user’s session with the application.
Although cookies often simplify the Internet shopping experience for the user,
they are controversial because they can also be considered as an invasion of privacy.
As we just saw, using a combination of cookies and user-supplied account information, a Web site can learn a lot about a user and potentially sell this information to a
third party. Cookie Central [Cookie Central 2012] includes extensive information
on the cookie controversy.
2.2.5 Web Caching
A Web cache—also called a proxy server—is a network entity that satisfies HTTP
requests on the behalf of an origin Web server. The Web cache has its own disk storage
and keeps copies of recently requested objects in this storage. As shown in Figure 2.11, a
user’s browser can be configured so that all of the user’s HTTP requests are first directed
to the Web cache. Once a browser is configured, each browser request for an object is
first directed to the Web cache. As an example, suppose a browser is requesting the
object http://www.someschool.edu/campus.gif. Here is what happens:
1. The browser establishes a TCP connection to the Web cache and sends an
HTTP request for the object to the Web cache.
2.2
HT
HT
Client
TP
TP
T
HT
Figure 2.11
res
T
ue
po
eq
Pr
HT
Client
req
ue
e
Pr
st
nse
st
spo
Proxy
server
HT
T
HT
HT
nse
eq
Pr
HT
T
es
Pr
TP
TP
ue
req
res
THE WEB AND HTTP
st
po
ue
po
•
nse
Origin
server
st
nse
Origin
server
Clients requesting objects through a Web cache
2. The Web cache checks to see if it has a copy of the object stored locally. If it
does, the Web cache returns the object within an HTTP response message to
the client browser.
3. If the Web cache does not have the object, the Web cache opens a TCP connection to the origin server, that is, to www.someschool.edu. The Web cache
then sends an HTTP request for the object into the cache-to-server TCP connection. After receiving this request, the origin server sends the object within
an HTTP response to the Web cache.
4. When the Web cache receives the object, it stores a copy in its local storage and
sends a copy, within an HTTP response message, to the client browser (over the
existing TCP connection between the client browser and the Web cache).
Note that a cache is both a server and a client at the same time. When it receives
requests from and sends responses to a browser, it is a server. When it sends requests
to and receives responses from an origin server, it is a client.
Typically a Web cache is purchased and installed by an ISP. For example, a university might install a cache on its campus network and configure all of the campus
browsers to point to the cache. Or a major residential ISP (such as AOL) might
install one or more caches in its network and preconfigure its shipped browsers to
point to the installed caches.
Web caching has seen deployment in the Internet for two reasons. First, a Web
cache can substantially reduce the response time for a client request, particularly if the
bottleneck bandwidth between the client and the origin server is much less than the bottleneck bandwidth between the client and the cache. If there is a high-speed connection
between the client and the cache, as there often is, and if the cache has the requested
object, then the cache will be able to deliver the object rapidly to the client. Second, as
we will soon illustrate with an example, Web caches can substantially reduce traffic on
111
112
CHAPTER 2
•
APPLICATION LAYER
Origin servers
Public Internet
15 Mbps access link
100 Mbps LAN
Institutional network
Figure 2.12
Bottleneck between an institutional network and the Internet
an institution’s access link to the Internet. By reducing traffic, the institution (for example, a company or a university) does not have to upgrade bandwidth as quickly, thereby
reducing costs. Furthermore, Web caches can substantially reduce Web traffic in the
Internet as a whole, thereby improving performance for all applications.
To gain a deeper understanding of the benefits of caches, let’s consider an example in the context of Figure 2.12. This figure shows two networks—the institutional
network and the rest of the public Internet. The institutional network is a high-speed
LAN. A router in the institutional network and a router in the Internet are connected
by a 15 Mbps link. The origin servers are attached to the Internet but are located all
over the globe. Suppose that the average object size is 1 Mbits and that the average
request rate from the institution’s browsers to the origin servers is 15 requests per
second. Suppose that the HTTP request messages are negligibly small and thus create no traffic in the networks or in the access link (from institutional router to Internet router). Also suppose that the amount of time it takes from when the router on the
Internet side of the access link in Figure 2.12 forwards an HTTP request (within an
IP datagram) until it receives the response (typically within many IP datagrams) is
two seconds on average. Informally, we refer to this last delay as the “Internet delay.”
2.2
•
THE WEB AND HTTP
The total response time—that is, the time from the browser’s request of an object
until its receipt of the object—is the sum of the LAN delay, the access delay (that is,
the delay between the two routers), and the Internet delay. Let’s now do a very crude
calculation to estimate this delay. The traffic intensity on the LAN (see Section 1.4.2) is
(15 requests/sec) ⭈ (1 Mbits/request)/(100 Mbps) = 0.15
whereas the traffic intensity on the access link (from the Internet router to institution
router) is
(15 requests/sec) ⭈ (1 Mbits/request)/(15 Mbps) = 1
A traffic intensity of 0.15 on a LAN typically results in, at most, tens of milliseconds of delay; hence, we can neglect the LAN delay. However, as discussed in
Section 1.4.2, as the traffic intensity approaches 1 (as is the case of the access link
in Figure 2.12), the delay on a link becomes very large and grows without bound.
Thus, the average response time to satisfy requests is going to be on the order of
minutes, if not more, which is unacceptable for the institution’s users. Clearly something must be done.
One possible solution is to increase the access rate from 15 Mbps to, say, 100
Mbps. This will lower the traffic intensity on the access link to 0.15, which translates to negligible delays between the two routers. In this case, the total response
time will roughly be two seconds, that is, the Internet delay. But this solution also
means that the institution must upgrade its access link from 15 Mbps to 100 Mbps, a
costly proposition.
Now consider the alternative solution of not upgrading the access link but
instead installing a Web cache in the institutional network. This solution is illustrated in Figure 2.13. Hit rates—the fraction of requests that are satisfied by a
cache—typically range from 0.2 to 0.7 in practice. For illustrative purposes, let’s
suppose that the cache provides a hit rate of 0.4 for this institution. Because the
clients and the cache are connected to the same high-speed LAN, 40 percent of
the requests will be satisfied almost immediately, say, within 10 milliseconds, by the
cache. Nevertheless, the remaining 60 percent of the requests still need to be satisfied by the origin servers. But with only 60 percent of the requested objects passing
through the access link, the traffic intensity on the access link is reduced from 1.0 to
0.6. Typically, a traffic intensity less than 0.8 corresponds to a small delay, say, tens
of milliseconds, on a 15 Mbps link. This delay is negligible compared with the twosecond Internet delay. Given these considerations, average delay therefore is
0.4 ⭈ (0.01 seconds) + 0.6 ⭈ (2.01 seconds)
which is just slightly greater than 1.2 seconds. Thus, this second solution provides an
even lower response time than the first solution, and it doesn’t require the institution
to upgrade its link to the Internet. The institution does, of course, have to purchase
113
114
CHAPTER 2
•
APPLICATION LAYER
Origin servers
Public Internet
15 Mbps access link
100 Mbps LAN
Institutional network
Figure 2.13
Institutional
cache
Adding a cache to the institutional network
and install a Web cache. But this cost is low—many caches use public-domain software that runs on inexpensive PCs.
Through the use of Content Distribution Networks (CDNs), Web caches are
increasingly playing an important role in the Internet. A CDN company installs many
geographically distributed caches throughout the Internet, thereby localizing much of
the traffic. There are shared CDNs (such as Akamai and Limelight) and dedicated CDNs
(such as Google and Microsoft). We will discuss CDNs in more detail in Chapter 7.
2.2.6 The Conditional GET
Although caching can reduce user-perceived response times, it introduces a new problem—the copy of an object residing in the cache may be stale. In other words, the
object housed in the Web server may have been modified since the copy was cached
at the client. Fortunately, HTTP has a mechanism that allows a cache to verify that its
objects are up to date. This mechanism is called the conditional GET. An HTTP
2.2
•
THE WEB AND HTTP
request message is a so-called conditional GET message if (1) the request message
uses the GET method and (2) the request message includes an If-ModifiedSince: header line.
To illustrate how the conditional GET operates, let’s walk through an example.
First, on the behalf of a requesting browser, a proxy cache sends a request message
to a Web server:
GET /fruit/kiwi.gif HTTP/1.1
Host: www.exotiquecuisine.com
Second, the Web server sends a response message with the requested object to the
cache:
HTTP/1.1 200 OK
Date: Sat, 8 Oct 2011 15:39:29
Server: Apache/1.3.0 (Unix)
Last-Modified: Wed, 7 Sep 2011 09:23:24
Content-Type: image/gif
(data data data data data ...)
The cache forwards the object to the requesting browser but also caches the object
locally. Importantly, the cache also stores the last-modified date along with the
object. Third, one week later, another browser requests the same object via the
cache, and the object is still in the cache. Since this object may have been modified
at the Web server in the past week, the cache performs an up-to-date check by issuing a conditional GET. Specifically, the cache sends:
GET /fruit/kiwi.gif HTTP/1.1
Host: www.exotiquecuisine.com
If-modified-since: Wed, 7 Sep 2011 09:23:24
Note that the value of the If-modified-since: header line is exactly equal to
the value of the Last-Modified: header line that was sent by the server one
week ago. This conditional GET is telling the server to send the object only if the
object has been modified since the specified date. Suppose the object has not been
modified since 7 Sep 2011 09:23:24. Then, fourth, the Web server sends a response
message to the cache:
HTTP/1.1 304 Not Modified
Date: Sat, 15 Oct 2011 15:39:29
Server: Apache/1.3.0 (Unix)
(empty entity body)
115
116
CHAPTER 2
•
APPLICATION LAYER
We see that in response to the conditional GET, the Web server still sends a response
message but does not include the requested object in the response message. Including
the requested object would only waste bandwidth and increase user-perceived response
time, particularly if the object is large. Note that this last response message has 304
Not Modified in the status line, which tells the cache that it can go ahead and forward its (the proxy cache’s) cached copy of the object to the requesting browser.
This ends our discussion of HTTP, the first Internet protocol (an application-layer
protocol) that we’ve studied in detail. We’ve seen the format of HTTP messages and
the actions taken by the Web client and server as these messages are sent and received.
We’ve also studied a bit of the Web’s application infrastructure, including caches, cookies, and back-end databases, all of which are tied in some way to the HTTP protocol.
2.3 File Transfer: FTP
In a typical FTP session, the user is sitting in front of one host (the local host)
and wants to transfer files to or from a remote host. In order for the user to
access the remote account, the user must provide a user identification and a password. After providing this authorization information, the user can transfer files
from the local file system to the remote file system and vice versa. As shown in
Figure 2.14, the user interacts with FTP through an FTP user agent. The user first
provides the hostname of the remote host, causing the FTP client process in the
local host to establish a TCP connection with the FTP server process in the
remote host. The user then provides the user identification and password, which
are sent over the TCP connection as part of FTP commands. Once the server has
authorized the user, the user copies one or more files stored in the local file system into the remote file system (or vice versa).
FTP user
interface
FTP
client
File transfer
FTP
server
User
or host
Local file
system
Figure 2.14
Remote file
system
FTP moves files between local and remote file systems
2.3
•
FILE TRANSFER: FTP
TCP control connection port 21
TCP data connection port 20
FTP
client
Figure 2.15
FTP
server
Control and data connections
HTTP and FTP are both file transfer protocols and have many common characteristics; for example, they both run on top of TCP. However, the two application-layer
protocols have some important differences. The most striking difference is that FTP
uses two parallel TCP connections to transfer a file, a control connection and a data
connection. The control connection is used for sending control information between
the two hosts—information such as user identification, password, commands to
change remote directory, and commands to “put” and “get” files. The data connection
is used to actually send a file. Because FTP uses a separate control connection, FTP is
said to send its control information out-of-band. HTTP, as you recall, sends request
and response header lines into the same TCP connection that carries the transferred
file itself. For this reason, HTTP is said to send its control information in-band. In the
next section, we’ll see that SMTP, the main protocol for electronic mail, also sends
control information in-band. The FTP control and data connections are illustrated in
Figure 2.15.
When a user starts an FTP session with a remote host, the client side of FTP
(user) first initiates a control TCP connection with the server side (remote host) on
server port number 21. The client side of FTP sends the user identification and
password over this control connection. The client side of FTP also sends, over the
control connection, commands to change the remote directory. When the server
side receives a command for a file transfer over the control connection (either to,
or from, the remote host), the server side initiates a TCP data connection to the
client side. FTP sends exactly one file over the data connection and then closes the
data connection. If, during the same session, the user wants to transfer another file,
FTP opens another data connection. Thus, with FTP, the control connection
remains open throughout the duration of the user session, but a new data connection is created for each file transferred within a session (that is, the data connections are non-persistent).
Throughout a session, the FTP server must maintain state about the user. In particular, the server must associate the control connection with a specific user account,
and the server must keep track of the user’s current directory as the user wanders
about the remote directory tree. Keeping track of this state information for each
ongoing user session significantly constrains the total number of sessions that FTP
can maintain simultaneously. Recall that HTTP, on the other hand, is stateless—it
does not have to keep track of any user state.
117
118
CHAPTER 2
•
APPLICATION LAYER
2.3.1 FTP Commands and Replies
We end this section with a brief discussion of some of the more common FTP commands and replies. The commands, from client to server, and replies, from server to
client, are sent across the control connection in 7-bit ASCII format. Thus, like HTTP
commands, FTP commands are readable by people. In order to delineate successive
commands, a carriage return and line feed end each command. Each command consists of four uppercase ASCII characters, some with optional arguments. Some of
the more common commands are given below:
• USER username: Used to send the user identification to the server.
• PASS password: Used to send the user password to the server.
• LIST: Used to ask the server to send back a list of all the files in the current
remote directory. The list of files is sent over a (new and non-persistent) data
connection rather than the control TCP connection.
• RETR filename: Used to retrieve (that is, get) a file from the current directory of the remote host. This command causes the remote host to initiate a data
connection and to send the requested file over the data connection.
• STOR filename: Used to store (that is, put) a file into the current directory
of the remote host.
There is typically a one-to-one correspondence between the command that the
user issues and the FTP command sent across the control connection. Each command is followed by a reply, sent from server to client. The replies are three-digit
numbers, with an optional message following the number. This is similar in structure to the status code and phrase in the status line of the HTTP response message.
Some typical replies, along with their possible messages, are as follows:
•
•
•
•
331
125
425
452
Username OK, password required
Data connection already open; transfer starting
Can’t open data connection
Error writing file
Readers who are interested in learning about the other FTP commands and replies
are encouraged to read RFC 959.
2.4 Electronic Mail in the Internet
Electronic mail has been around since the beginning of the Internet. It was the most
popular application when the Internet was in its infancy [Segaller 1998], and has
2.4
•
ELECTRONIC MAIL IN THE INTERNET
become more and more elaborate and powerful over the years. It remains one of the
Internet’s most important and utilized applications.
As with ordinary postal mail, e-mail is an asynchronous communication
medium—people send and read messages when it is convenient for them, without
having to coordinate with other people’s schedules. In contrast with postal mail, electronic mail is fast, easy to distribute, and inexpensive. Modern e-mail has many powerful features, including messages with attachments, hyperlinks, HTML-formatted
text, and embedded photos.
In this section, we examine the application-layer protocols that are at the heart
of Internet e-mail. But before we jump into an in-depth discussion of these protocols, let’s take a high-level view of the Internet mail system and its key components.
Figure 2.16 presents a high-level view of the Internet mail system. We see from
this diagram that it has three major components: user agents, mail servers, and the
Mail server
User agent
SM
SMTP
TP
SM
TP
Mail server
User agent
Mail server
User agent
User agent
User agent
User agent
Key:
Outgoing
message queue
Figure 2.16
User mailbox
A high-level view of the Internet e-mail system
119
120
CHAPTER 2
•
APPLICATION LAYER
Simple Mail Transfer Protocol (SMTP). We now describe each of these components in the context of a sender, Alice, sending an e-mail message to a recipient,
Bob. User agents allow users to read, reply to, forward, save, and compose messages. Microsoft Outlook and Apple Mail are examples of user agents for e-mail.
When Alice is finished composing her message, her user agent sends the message to
her mail server, where the message is placed in the mail server’s outgoing message
queue. When Bob wants to read a message, his user agent retrieves the message
from his mailbox in his mail server.
Mail servers form the core of the e-mail infrastructure. Each recipient, such as
Bob, has a mailbox located in one of the mail servers. Bob’s mailbox manages and
maintains the messages that have been sent to him. A typical message starts its journey in the sender’s user agent, travels to the sender’s mail server, and
travels to the recipient’s mail server, where it is deposited in the recipient’s mailbox.
CASE HISTORY
WEB E-MAIL
In December 1995, just a few years after the Web was “invented,” Sabeer Bhatia
and Jack Smith visited the Internet venture capitalist Draper Fisher Jurvetson and
proposed developing a free Web-based e-mail system. The idea was to give a free
e-mail account to anyone who wanted one, and to make the accounts accessible
from the Web. In exchange for 15 percent of the company, Draper Fisher
Jurvetson financed Bhatia and Smith, who formed a company called Hotmail.
With three full-time people and 14 part-time people who worked for stock options,
they were able to develop and launch the service in July 1996. Within a month
after launch, they had 100,000 subscribers. In December 1997, less than 18
months after launching the service, Hotmail had over 12 million subscribers and
was acquired by Microsoft, reportedly for $400 million. The success of Hotmail is
often attributed to its “first-mover advantage” and to the intrinsic “viral marketing”
of e-mail. (Perhaps some of the students reading this book will be among the new
entrepreneurs who conceive and develop first-mover Internet services with inherent
viral marketing.)
Web e-mail continues to thrive, becoming more sophisticated and powerful every
year. One of the most popular services today is Google’s gmail, which offers gigabytes of free storage, advanced spam filtering and virus detection, e-mail encryption
(using SSL), mail fetching from third-party e-mail services, and a search-oriented interface. Asynchronous messaging within social networks, such as Facebook, has also
become popular in recent years.
2.4
•
ELECTRONIC MAIL IN THE INTERNET
When Bob wants to access the messages in his mailbox, the mail server
containing his mailbox authenticates Bob (with usernames and passwords). Alice’s
mail server must also deal with failures in Bob’s mail server. If Alice’s server cannot deliver mail to Bob’s server, Alice’s server holds the message in a message
queue and attempts to transfer the message later. Reattempts are often done every
30 minutes or so; if there is no success after several days, the server removes the
message and notifies the sender (Alice) with an e-mail message.
SMTP is the principal application-layer protocol for Internet electronic mail. It
uses the reliable data transfer service of TCP to transfer mail from the sender’s mail
server to the recipient’s mail server. As with most application-layer protocols,
SMTP has two sides: a client side, which executes on the sender’s mail server, and a
server side, which executes on the recipient’s mail server. Both the client and server
sides of SMTP run on every mail server. When a mail server sends mail to other
mail servers, it acts as an SMTP client. When a mail server receives mail from other
mail servers, it acts as an SMTP server.
2.4.1 SMTP
SMTP, defined in RFC 5321, is at the heart of Internet electronic mail. As mentioned above, SMTP transfers messages from senders’ mail servers to the recipients’ mail servers. SMTP is much older than HTTP. (The original SMTP RFC
dates back to 1982, and SMTP was around long before that.) Although SMTP has
numerous wonderful qualities, as evidenced by its ubiquity in the Internet, it is
nevertheless a legacy technology that possesses certain archaic characteristics.
For example, it restricts the body (not just the headers) of all mail messages to
simple 7-bit ASCII. This restriction made sense in the early 1980s when transmission capacity was scarce and no one was e-mailing large attachments or large
image, audio, or video files. But today, in the multimedia era, the 7-bit ASCII
restriction is a bit of a pain—it requires binary multimedia data to be encoded to
ASCII before being sent over SMTP; and it requires the corresponding ASCII
message to be decoded back to binary after SMTP transport. Recall from Section
2.2 that HTTP does not require multimedia data to be ASCII encoded before
transfer.
To illustrate the basic operation of SMTP, let’s walk through a common scenario. Suppose Alice wants to send Bob a simple ASCII message.
1. Alice invokes her user agent for e-mail, provides Bob’s e-mail address (for
example, [email protected]), composes a message, and instructs the
user agent to send the message.
2. Alice’s user agent sends the message to her mail server, where it is placed in a
message queue.
121
CHAPTER 2
122
•
APPLICATION LAYER
Alice’s
mail server
1
Alice’s
agent
3
2
Bob’s
mail server
4
SMTP
6
Bob’s
agent
5
Key:
Message queue
User mailbox
Figure 2.17
Alice sends a message to Bob
3. The client side of SMTP, running on Alice’s mail server, sees the message in
the message queue. It opens a TCP connection to an SMTP server, running on
Bob’s mail server.
4. After some initial SMTP handshaking, the SMTP client sends Alice’s message
into the TCP connection.
5. At Bob’s mail server, the server side of SMTP receives the message. Bob’s
mail server then places the message in Bob’s mailbox.
6. Bob invokes his user agent to read the message at his convenience.
The scenario is summarized in Figure 2.17.
It is important to observe that SMTP does not normally use intermediate mail
servers for sending mail, even when the two mail servers are located at opposite
ends of the world. If Alice’s server is in Hong Kong and Bob’s server is in St. Louis,
the TCP connection is a direct connection between the Hong Kong and St. Louis
servers. In particular, if Bob’s mail server is down, the message remains in Alice’s
mail server and waits for a new attempt—the message does not get placed in some
intermediate mail server.
Let’s now take a closer look at how SMTP transfers a message from a sending mail server to a receiving mail server. We will see that the SMTP protocol has
many similarities with protocols that are used for face-to-face human interaction.
First, the client SMTP (running on the sending mail server host) has TCP establish a connection to port 25 at the server SMTP (running on the receiving mail
server host). If the server is down, the client tries again later. Once this connection is established, the server and client perform some application-layer
handshaking—just as humans often introduce themselves before transferring
information from one to another, SMTP clients and servers introduce themselves
before transferring information. During this SMTP handshaking phase, the
2.4
•
ELECTRONIC MAIL IN THE INTERNET
SMTP client indicates the e-mail address of the sender (the person who generated
the message) and the e-mail address of the recipient. Once the SMTP client and
server have introduced themselves to each other, the client sends the message.
SMTP can count on the reliable data transfer service of TCP to get the message
to the server without errors. The client then repeats this process over the same
TCP connection if it has other messages to send to the server; otherwise, it
instructs TCP to close the connection.
Let’s next take a look at an example transcript of messages exchanged
between an SMTP client (C) and an SMTP server (S). The hostname of the client
is crepes.fr and the hostname of the server is hamburger.edu. The
ASCII text lines prefaced with C: are exactly the lines the client sends into its
TCP socket, and the ASCII text lines prefaced with S: are exactly the lines the
server sends into its TCP socket. The following transcript begins as soon as the
TCP connection is established.
S:
C:
S:
C:
S:
C:
S:
C:
S:
C:
C:
C:
S:
C:
S:
220 hamburger.edu
HELO crepes.fr
250 Hello crepes.fr, pleased to meet you
MAIL FROM: <[email protected]>
250 [email protected] ... Sender ok
RCPT TO: <[email protected]>
250 [email protected] ... Recipient ok
DATA
354 Enter mail, end with “.” on a line by itself
Do you like ketchup?
How about pickles?
.
250 Message accepted for delivery
QUIT
221 hamburger.edu closing connection
In the example above, the client sends a message (“Do you like ketchup?
How about pickles?”) from mail server crepes.fr to mail server hamburger.edu. As part of the dialogue, the client issued five commands: HELO (an
abbreviation for HELLO), MAIL FROM, RCPT TO, DATA, and QUIT. These commands are self-explanatory. The client also sends a line consisting of a single period,
which indicates the end of the message to the server. (In ASCII jargon, each message ends with CRLF.CRLF, where CR and LF stand for carriage return and line
feed, respectively.) The server issues replies to each command, with each reply having a reply code and some (optional) English-language explanation. We mention
here that SMTP uses persistent connections: If the sending mail server has several
messages to send to the same receiving mail server, it can send all of the messages
over the same TCP connection. For each message, the client begins the process with
123
124
CHAPTER 2
•
APPLICATION LAYER
a new MAIL FROM: crepes.fr, designates the end of message with an isolated
period, and issues QUIT only after all messages have been sent.
It is highly recommended that you use Telnet to carry out a direct dialogue with
an SMTP server. To do this, issue
telnet serverName 25
where serverName is the name of a local mail server. When you do this, you are
simply establishing a TCP connection between your local host and the mail server.
After typing this line, you should immediately receive the 220 reply from the
server. Then issue the SMTP commands HELO, MAIL FROM, RCPT TO, DATA,
CRLF.CRLF, and QUIT at the appropriate times. It is also highly recommended
that you do Programming Assignment 3 at the end of this chapter. In that assignment, you’ll build a simple user agent that implements the client side of SMTP.
It will allow you to send an e-mail message to an arbitrary recipient via a local
mail server.
2.4.2 Comparison with HTTP
Let’s now briefly compare SMTP with HTTP. Both protocols are used to transfer
files from one host to another: HTTP transfers files (also called objects) from a Web
server to a Web client (typically a browser); SMTP transfers files (that is, e-mail
messages) from one mail server to another mail server. When transferring the files,
both persistent HTTP and SMTP use persistent connections. Thus, the two protocols
have common characteristics. However, there are important differences. First,
HTTP is mainly a pull protocol—someone loads information on a Web server and
users use HTTP to pull the information from the server at their convenience. In particular, the TCP connection is initiated by the machine that wants to receive the file.
On the other hand, SMTP is primarily a push protocol—the sending mail server
pushes the file to the receiving mail server. In particular, the TCP connection is initiated by the machine that wants to send the file.
A second difference, which we alluded to earlier, is that SMTP requires
each message, including the body of each message, to be in 7-bit ASCII format.
If the message contains characters that are not 7-bit ASCII (for example, French
characters with accents) or contains binary data (such as an image file), then the
message has to be encoded into 7-bit ASCII. HTTP data does not impose this
restriction.
A third important difference concerns how a document consisting of text and
images (along with possibly other media types) is handled. As we learned in Section
2.2, HTTP encapsulates each object in its own HTTP response message. Internet
mail places all of the message’s objects into one message.
2.4
•
ELECTRONIC MAIL IN THE INTERNET
2.4.3 Mail Message Formats
When Alice writes an ordinary snail-mail letter to Bob, she may include all kinds of
peripheral header information at the top of the letter, such as Bob’s address, her own
return address, and the date. Similarly, when an e-mail message is sent from one person to another, a header containing peripheral information precedes the body of the
message itself. This peripheral information is contained in a series of header lines,
which are defined in RFC 5322. The header lines and the body of the message are
separated by a blank line (that is, by CRLF). RFC 5322 specifies the exact format for
mail header lines as well as their semantic interpretations. As with HTTP, each header
line contains readable text, consisting of a keyword followed by a colon followed by
a value. Some of the keywords are required and others are optional. Every header
must have a From: header line and a To: header line; a header may include a Subject: header line as well as other optional header lines. It is important to note that
these header lines are different from the SMTP commands we studied in Section 2.4.1
(even though they contain some common words such as “from” and “to”). The commands in that section were part of the SMTP handshaking protocol; the header lines
examined in this section are part of the mail message itself.
A typical message header looks like this:
From: [email protected]
To: [email protected]
Subject: Searching for the meaning of life.
After the message header, a blank line follows; then the message body (in ASCII)
follows. You should use Telnet to send a message to a mail server that contains some
header lines, including the Subject: header line. To do this, issue telnet
serverName 25, as discussed in Section 2.4.1.
2.4.4 Mail Access Protocols
Once SMTP delivers the message from Alice’s mail server to Bob’s mail server, the
message is placed in Bob’s mailbox. Throughout this discussion we have tacitly
assumed that Bob reads his mail by logging onto the server host and then executing
a mail reader that runs on that host. Up until the early 1990s this was the standard
way of doing things. But today, mail access uses a client-server architecture—the
typical user reads e-mail with a client that executes on the user’s end system, for
example, on an office PC, a laptop, or a smartphone. By executing a mail client on a
local PC, users enjoy a rich set of features, including the ability to view multimedia
messages and attachments.
Given that Bob (the recipient) executes his user agent on his local PC, it is natural to consider placing a mail server on his local PC as well. With this approach,
125
126
CHAPTER 2
•
APPLICATION LAYER
Alice’s mail server would dialogue directly with Bob’s PC. There is a problem with
this approach, however. Recall that a mail server manages mailboxes and runs the
client and server sides of SMTP. If Bob’s mail server were to reside on his local PC,
then Bob’s PC would have to remain always on, and connected to the Internet, in
order to receive new mail, which can arrive at any time. This is impractical for many
Internet users. Instead, a typical user runs a user agent on the local PC but accesses
its mailbox stored on an always-on shared mail server. This mail server is shared
with other users and is typically maintained by the user’s ISP (for example, university or company).
Now let’s consider the path an e-mail message takes when it is sent from Alice
to Bob. We just learned that at some point along the path the e-mail message needs
to be deposited in Bob’s mail server. This could be done simply by having Alice’s
user agent send the message directly to Bob’s mail server. And this could be done
with SMTP—indeed, SMTP has been designed for pushing e-mail from one host to
another. However, typically the sender’s user agent does not dialogue directly with
the recipient’s mail server. Instead, as shown in Figure 2.18, Alice’s user agent uses
SMTP to push the e-mail message into her mail server, then Alice’s mail server uses
SMTP (as an SMTP client) to relay the e-mail message to Bob’s mail server. Why
the two-step procedure? Primarily because without relaying through Alice’s mail
server, Alice’s user agent doesn’t have any recourse to an unreachable destination
mail server. By having Alice first deposit the e-mail in her own mail server, Alice’s
mail server can repeatedly try to send the message to Bob’s mail server, say every
30 minutes, until Bob’s mail server becomes operational. (And if Alice’s mail server
is down, then she has the recourse of complaining to her system administrator!) The
SMTP RFC defines how the SMTP commands can be used to relay a message
across multiple SMTP servers.
But there is still one missing piece to the puzzle! How does a recipient like Bob,
running a user agent on his local PC, obtain his messages, which are sitting in a mail
server within Bob’s ISP? Note that Bob’s user agent can’t use SMTP to obtain the
messages because obtaining the messages is a pull operation, whereas SMTP is a
Alice’s
mail server
Alice’s
agent
SMTP
Figure 2.18
Bob’s
mail server
SMTP
POP3,
IMAP, or
HTTP
Bob’s
agent
E-mail protocols and their communicating entities
2.4
•
ELECTRONIC MAIL IN THE INTERNET
push protocol. The puzzle is completed by introducing a special mail access protocol that transfers messages from Bob’s mail server to his local PC. There are currently a number of popular mail access protocols, including Post Office
Protocol—Version 3 (POP3), Internet Mail Access Protocol (IMAP), and HTTP.
Figure 2.18 provides a summary of the protocols that are used for Internet mail:
SMTP is used to transfer mail from the sender’s mail server to the recipient’s mail
server; SMTP is also used to transfer mail from the sender’s user agent to the
sender’s mail server. A mail access protocol, such as POP3, is used to transfer mail
from the recipient’s mail server to the recipient’s user agent.
POP3
POP3 is an extremely simple mail access protocol. It is defined in [RFC 1939], which
is short and quite readable. Because the protocol is so simple, its functionality is
rather limited. POP3 begins when the user agent (the client) opens a TCP connection to the mail server (the server) on port 110. With the TCP connection established, POP3 progresses through three phases: authorization, transaction, and update.
During the first phase, authorization, the user agent sends a username and a password
(in the clear) to authenticate the user. During the second phase, transaction, the user
agent retrieves messages; also during this phase, the user agent can mark messages
for deletion, remove deletion marks, and obtain mail statistics. The third phase,
update, occurs after the client has issued the quit command, ending the POP3
session; at this time, the mail server deletes the messages that were marked for
deletion.
In a POP3 transaction, the user agent issues commands, and the server responds
to each command with a reply. There are two possible responses: +OK (sometimes
followed by server-to-client data), used by the server to indicate that the previous
command was fine; and -ERR, used by the server to indicate that something was
wrong with the previous command.
The authorization phase has two principal commands: user <username> and
pass <password>. To illustrate these two commands, we suggest that you Telnet
directly into a POP3 server, using port 110, and issue these commands. Suppose that
mailServer is the name of your mail server. You will see something like:
telnet mailServer 110
+OK POP3 server ready
user bob
+OK
pass hungry
+OK user successfully logged on
If you misspell a command, the POP3 server will reply with an -ERR message.
127
128
CHAPTER 2
•
APPLICATION LAYER
Now let’s take a look at the transaction phase. A user agent using POP3 can
often be configured (by the user) to “download and delete” or to “download and
keep.” The sequence of commands issued by a POP3 user agent depends on which
of these two modes the user agent is operating in. In the download-and-delete mode,
the user agent will issue the list, retr, and dele commands. As an example,
suppose the user has two messages in his or her mailbox. In the dialogue below, C:
(standing for client) is the user agent and S: (standing for server) is the mail server.
The transaction will look something like:
C:
S:
S:
S:
C:
S:
S:
S:
S:
C:
C:
S:
S:
S:
S:
C:
C:
S:
list
1 498
2 912
.
retr 1
(blah blah ...
.................
..........blah)
.
dele 1
retr 2
(blah blah ...
.................
..........blah)
.
dele 2
quit
+OK POP3 server signing off
The user agent first asks the mail server to list the size of each of the stored messages. The user agent then retrieves and deletes each message from the server. Note
that after the authorization phase, the user agent employed only four commands:
list, retr, dele, and quit. The syntax for these commands is defined in RFC
1939. After processing the quit command, the POP3 server enters the update
phase and removes messages 1 and 2 from the mailbox.
A problem with this download-and-delete mode is that the recipient, Bob, may
be nomadic and may want to access his mail messages from multiple machines, for
example, his office PC, his home PC, and his portable computer. The downloadand-delete mode partitions Bob’s mail messages over these three machines; in particular, if Bob first reads a message on his office PC, he will not be able to reread
the message from his portable at home later in the evening. In the download-andkeep mode, the user agent leaves the messages on the mail server after downloading
them. In this case, Bob can reread messages from different machines; he can access
a message from work and access it again later in the week from home.
2.4
•
ELECTRONIC MAIL IN THE INTERNET
During a POP3 session between a user agent and the mail server, the POP3
server maintains some state information; in particular, it keeps track of which user
messages have been marked deleted. However, the POP3 server does not carry state
information across POP3 sessions. This lack of state information across sessions
greatly simplifies the implementation of a POP3 server.
IMAP
With POP3 access, once Bob has downloaded his messages to the local machine,
he can create mail folders and move the downloaded messages into the folders.
Bob can then delete messages, move messages across folders, and search for
messages (by sender name or subject). But this paradigm—namely, folders and
messages in the local machine—poses a problem for the nomadic user, who
would prefer to maintain a folder hierarchy on a remote server that can be
accessed from any computer. This is not possible with POP3—the POP3 protocol
does not provide any means for a user to create remote folders and assign messages to folders.
To solve this and other problems, the IMAP protocol, defined in [RFC 3501],
was invented. Like POP3, IMAP is a mail access protocol. It has many more features than POP3, but it is also significantly more complex. (And thus the client and
server side implementations are significantly more complex.)
An IMAP server will associate each message with a folder; when a message first
arrives at the server, it is associated with the recipient’s INBOX folder. The recipient
can then move the message into a new, user-created folder, read the message, delete
the message, and so on. The IMAP protocol provides commands to allow users to
create folders and move messages from one folder to another. IMAP also provides
commands that allow users to search remote folders for messages matching specific
criteria. Note that, unlike POP3, an IMAP server maintains user state information
across IMAP sessions—for example, the names of the folders and which messages
are associated with which folders.
Another important feature of IMAP is that it has commands that permit a user
agent to obtain components of messages. For example, a user agent can obtain just
the message header of a message or just one part of a multipart MIME message.
This feature is useful when there is a low-bandwidth connection (for example, a
slow-speed modem link) between the user agent and its mail server. With a lowbandwidth connection, the user may not want to download all of the messages in
its mailbox, particularly avoiding long messages that might contain, for example,
an audio or video clip.
Web-Based E-Mail
More and more users today are sending and accessing their e-mail through their Web
browsers. Hotmail introduced Web-based access in the mid 1990s. Now Web-based
129
130
CHAPTER 2
•
APPLICATION LAYER
e-mail is also provided by Google, Yahoo!, as well as just about every major university and corporation. With this service, the user agent is an ordinary Web browser,
and the user communicates with its remote mailbox via HTTP. When a recipient,
such as Bob, wants to access a message in his mailbox, the e-mail message is sent
from Bob’s mail server to Bob’s browser using the HTTP protocol rather than the
POP3 or IMAP protocol. When a sender, such as Alice, wants to send an e-mail
message, the e-mail message is sent from her browser to her mail server over HTTP
rather than over SMTP. Alice’s mail server, however, still sends messages to, and
receives messages from, other mail servers using SMTP.
2.5 DNS—The Internet’s Directory Service
We human beings can be identified in many ways. For example, we can be identified by the names that appear on our birth certificates. We can be identified by our
social security numbers. We can be identified by our driver’s license numbers.
Although each of these identifiers can be used to identify people, within a given
context one identifier may be more appropriate than another. For example, the computers at the IRS (the infamous tax-collecting agency in the United States) prefer to
use fixed-length social security numbers rather than birth certificate names. On the
other hand, ordinary people prefer the more mnemonic birth certificate names rather
than social security numbers. (Indeed, can you imagine saying, “Hi. My name is
132-67-9875. Please meet my husband, 178-87-1146.”)
Just as humans can be identified in many ways, so too can Internet hosts. One identifier for a host is its hostname. Hostnames—such as cnn.com, www.yahoo.
com, gaia.cs.umass.edu, and cis.poly.edu—are mnemonic and are therefore appreciated by humans. However, hostnames provide little, if any, information about
the location within the Internet of the host. (A hostname such as www.eurecom.fr,
which ends with the country code .fr, tells us that the host is probably in France, but
doesn’t say much more.) Furthermore, because hostnames can consist of variablelength alphanumeric characters, they would be difficult to process by routers. For these
reasons, hosts are also identified by so-called IP addresses.
We discuss IP addresses in some detail in Chapter 4, but it is useful to say a few
brief words about them now. An IP address consists of four bytes and has a rigid
hierarchical structure. An IP address looks like 121.7.106.83, where each
period separates one of the bytes expressed in decimal notation from 0 to 255. An IP
address is hierarchical because as we scan the address from left to right, we obtain
more and more specific information about where the host is located in the Internet
(that is, within which network, in the network of networks). Similarly, when we scan
a postal address from bottom to top, we obtain more and more specific information
about where the addressee is located.
2.5
•
DNS—THE INTERNET’S DIRECTORY SERVICE
2.5.1 Services Provided by DNS
We have just seen that there are two ways to identify a host—by a hostname and by
an IP address. People prefer the more mnemonic hostname identifier, while routers
prefer fixed-length, hierarchically structured IP addresses. In order to reconcile
these preferences, we need a directory service that translates hostnames to IP
addresses. This is the main task of the Internet’s domain name system (DNS). The
DNS is (1) a distributed database implemented in a hierarchy of DNS servers, and
(2) an application-layer protocol that allows hosts to query the distributed database.
The DNS servers are often UNIX machines running the Berkeley Internet Name
Domain (BIND) software [BIND 2012]. The DNS protocol runs over UDP and uses
port 53.
DNS is commonly employed by other application-layer protocols—including
HTTP, SMTP, and FTP—to translate user-supplied hostnames to IP addresses. As
an example, consider what happens when a browser (that is, an HTTP client),
running on some user’s host, requests the URL www.someschool.edu/
index.html. In order for the user’s host to be able to send an HTTP request message to the Web server www.someschool.edu, the user’s host must first obtain
the IP address of www.someschool.edu. This is done as follows.
1. The same user machine runs the client side of the DNS application.
2. The browser extracts the hostname, www.someschool.edu, from the URL
and passes the hostname to the client side of the DNS application.
3. The DNS client sends a query containing the hostname to a DNS server.
4. The DNS client eventually receives a reply, which includes the IP address for
the hostname.
5. Once the browser receives the IP address from DNS, it can initiate a TCP connection to the HTTP server process located at port 80 at that IP address.
We see from this example that DNS adds an additional delay—sometimes substantial—to the Internet applications that use it. Fortunately, as we discuss below, the
desired IP address is often cached in a “nearby” DNS server, which helps to reduce
DNS network traffic as well as the average DNS delay.
DNS provides a few other important services in addition to translating hostnames to IP addresses:
• Host aliasing. A host with a complicated hostname can have one or more alias
names. For example, a hostname such as relay1.west-coast.enterprise.com could have, say, two aliases such as enterprise.com and
www.enterprise.com. In this case, the hostname relay1.westcoast.enterprise.com is said to be a canonical hostname. Alias hostnames, when present, are typically more mnemonic than canonical hostnames.
131
132
CHAPTER 2
•
APPLICATION LAYER
PRINCIPLES IN PRACTICE
DNS: CRITICAL NETWORK FUNCTIONS VIA THE CLIENT-SERVER
PARADIGM
Like HTTP, FTP, and SMTP, the DNS protocol is an application-layer protocol since it (1) runs
between communicating end systems using the client-server paradigm and (2) relies on an
underlying end-to-end transport protocol to transfer DNS messages between communicating
end systems. In another sense, however, the role of the DNS is quite different from Web,
file transfer, and e-mail applications. Unlike these applications, the DNS is not an application with which a user directly interacts. Instead, the DNS provides a core Internet function—namely, translating hostnames to their underlying IP addresses, for user applications
and other software in the Internet. We noted in Section 1.2 that much of the complexity in
the Internet architecture is located at the “edges” of the network. The DNS, which implements the critical name-to-address translation process using clients and servers located at
the edge of the network, is yet another example of that design philosophy.
DNS can be invoked by an application to obtain the canonical hostname for a
supplied alias hostname as well as the IP address of the host.
• Mail server aliasing. For obvious reasons, it is highly desirable that e-mail
addresses be mnemonic. For example, if Bob has an account with Hotmail, Bob’s
e-mail address might be as simple as [email protected] However, the hostname of the Hotmail mail server is more complicated and much less mnemonic
than simply hotmail.com (for example, the canonical hostname might be
something like relay1.west-coast.hotmail.com). DNS can be
invoked by a mail application to obtain the canonical hostname for a supplied
alias hostname as well as the IP address of the host. In fact, the MX record (see
below) permits a company’s mail server and Web server to have identical
(aliased) hostnames; for example, a company’s Web server and mail server can
both be called enterprise.com.
• Load distribution. DNS is also used to perform load distribution among replicated servers, such as replicated Web servers. Busy sites, such as cnn.com, are
replicated over multiple servers, with each server running on a different end system and each having a different IP address. For replicated Web servers, a set of
IP addresses is thus associated with one canonical hostname. The DNS database
contains this set of IP addresses. When clients make a DNS query for a name
mapped to a set of addresses, the server responds with the entire set of IP
addresses, but rotates the ordering of the addresses within each reply. Because a
client typically sends its HTTP request message to the IP address that is listed
first in the set, DNS rotation distributes the traffic among the replicated servers.
2.5
•
DNS—THE INTERNET’S DIRECTORY SERVICE
DNS rotation is also used for e-mail so that multiple mail servers can have the
same alias name. Also, content distribution companies such as Akamai have used
DNS in more sophisticated ways [Dilley 2002] to provide Web content distribution (see Chapter 7).
The DNS is specified in RFC 1034 and RFC 1035, and updated in several
additional RFCs. It is a complex system, and we only touch upon key aspects of
its operation here. The interested reader is referred to these RFCs and the book
by Albitz and Liu [Albitz 1993]; see also the retrospective paper [Mockapetris
1988], which provides a nice description of the what and why of DNS, and
[Mockapetris 2005].
2.5.2 Overview of How DNS Works
We now present a high-level overview of how DNS works. Our discussion will
focus on the hostname-to-IP-address translation service.
Suppose that some application (such as a Web browser or a mail reader) running in a user’s host needs to translate a hostname to an IP address. The application will invoke the client side of DNS, specifying the hostname that needs to be
translated. (On many UNIX-based machines, gethostbyname() is the function call that an application calls in order to perform the translation.) DNS in the
user’s host then takes over, sending a query message into the network. All DNS
query and reply messages are sent within UDP datagrams to port 53. After a delay,
ranging from milliseconds to seconds, DNS in the user’s host receives a DNS
reply message that provides the desired mapping. This mapping is then passed to
the invoking application. Thus, from the perspective of the invoking application
in the user’s host, DNS is a black box providing a simple, straightforward translation service. But in fact, the black box that implements the service is complex,
consisting of a large number of DNS servers distributed around the globe, as well
as an application-layer protocol that specifies how the DNS servers and querying
hosts communicate.
A simple design for DNS would have one DNS server that contains all the mappings. In this centralized design, clients simply direct all queries to the single DNS
server, and the DNS server responds directly to the querying clients. Although the
simplicity of this design is attractive, it is inappropriate for today’s Internet, with its
vast (and growing) number of hosts. The problems with a centralized design
include:
• A single point of failure. If the DNS server crashes, so does the entire Internet!
• Traffic volume. A single DNS server would have to handle all DNS queries (for
all the HTTP requests and e-mail messages generated from hundreds of millions
of hosts).
133
134
CHAPTER 2
•
APPLICATION LAYER
• Distant centralized database. A single DNS server cannot be “close to” all the
querying clients. If we put the single DNS server in New York City, then all
queries from Australia must travel to the other side of the globe, perhaps over
slow and congested links. This can lead to significant delays.
• Maintenance. The single DNS server would have to keep records for all Internet
hosts. Not only would this centralized database be huge, but it would have to be
updated frequently to account for every new host.
In summary, a centralized database in a single DNS server simply doesn’t scale.
Consequently, the DNS is distributed by design. In fact, the DNS is a wonderful
example of how a distributed database can be implemented in the Internet.
A Distributed, Hierarchical Database
In order to deal with the issue of scale, the DNS uses a large number of servers,
organized in a hierarchical fashion and distributed around the world. No single DNS
server has all of the mappings for all of the hosts in the Internet. Instead, the mappings are distributed across the DNS servers. To a first approximation, there are
three classes of DNS servers—root DNS servers, top-level domain (TLD) DNS
servers, and authoritative DNS servers—organized in a hierarchy as shown in Figure 2.19. To understand how these three classes of servers interact, suppose a DNS
client wants to determine the IP address for the hostname www.amazon.com. To
a first approximation, the following events will take place. The client first contacts
one of the root servers, which returns IP addresses for TLD servers for the top-level
domain com. The client then contacts one of these TLD servers, which returns the
IP address of an authoritative server for amazon.com. Finally, the client contacts
one of the authoritative servers for amazon.com, which returns the IP address
Root DNS servers
com DNS servers
yahoo.com
DNS servers
Figure 2.19
amazon.com
DNS servers
org DNS servers
pbs.org
DNS servers
edu DNS servers
poly.edu
DNS servers
Portion of the hierarchy of DNS servers
umass.edu
DNS servers
2.5
c.
d.
h.
j.
e. NASA Mt View, CA
f. Internet Software C.
Palo Alto, CA
(and 48 other sites)
•
Cogent, Herndon, VA (5 other sites)
U Maryland College Park, MD
ARL Aberdeen, MD
Verisign, Dulles VA (69 other sites )
g. US DoD Columbus, OH
(5 other sites)
a. Verisign, Los Angeles CA
(5 other sites)
b. USC-ISI Marina del Rey, CA
l. ICANN Los Angeles, CA
(41 other sites)
Figure 2.20
DNS—THE INTERNET’S DIRECTORY SERVICE
i. Netnod, Stockholm
(37 other sites)
k. RIPE London
(17 other sites)
m. WIDE Tokyo
(5 other sites)
DNS root servers in 2012 (name, organization, location)
for the hostname www.amazon.com. We’ll soon examine this DNS lookup
process in more detail. But let’s first take a closer look at these three classes of
DNS servers:
• Root DNS servers. In the Internet there are 13 root DNS servers (labeled A
through M), most of which are located in North America. An October 2006 map
of the root DNS servers is shown in Figure 2.20; a list of the current root DNS
servers is available via [Root-servers 2012]. Although we have referred to each
of the 13 root DNS servers as if it were a single server, each “server” is actually
a network of replicated servers, for both security and reliability purposes. All
together, there are 247 root servers as of fall 2011.
• Top-level domain (TLD) servers. These servers are responsible for top-level
domains such as com, org, net, edu, and gov, and all of the country top-level domains
such as uk, fr, ca, and jp. The company Verisign Global Registry Services
maintains the TLD servers for the com top-level domain, and the company
Educause maintains the TLD servers for the edu top-level domain. See [IANA
TLD 2012] for a list of all top-level domains.
• Authoritative DNS servers. Every organization with publicly accessible hosts
(such as Web servers and mail servers) on the Internet must provide publicly accessible DNS records that map the names of those hosts to IP addresses. An organization’s authoritative DNS server houses these DNS records. An organization can
135
136
CHAPTER 2
•
APPLICATION LAYER
choose to implement its own authoritative DNS server to hold these records; alternatively, the organization can pay to have these records stored in an authoritative
DNS server of some service provider. Most universities and large companies
implement and maintain their own primary and secondary (backup) authoritative
DNS server.
The root, TLD, and authoritative DNS servers all belong to the hierarchy of
DNS servers, as shown in Figure 2.19. There is another important type of DNS
server called the local DNS server. A local DNS server does not strictly belong to
the hierarchy of servers but is nevertheless central to the DNS architecture. Each
ISP—such as a university, an academic department, an employee’s company, or a
residential ISP—has a local DNS server (also called a default name server). When a
host connects to an ISP, the ISP provides the host with the IP addresses of one or
more of its local DNS servers (typically through DHCP, which is discussed in Chapter 4). You can easily determine the IP address of your local DNS server by accessing network status windows in Windows or UNIX. A host’s local DNS server is
typically “close to” the host. For an institutional ISP, the local DNS server may be
on the same LAN as the host; for a residential ISP, it is typically separated from the
host by no more than a few routers. When a host makes a DNS query, the query is
sent to the local DNS server, which acts a proxy, forwarding the query into the DNS
server hierarchy, as we’ll discuss in more detail below.
Let’s take a look at a simple example. Suppose the host cis.poly.edu
desires the IP address of gaia.cs.umass.edu. Also suppose that Polytechnic’s
local DNS server is called dns.poly.edu and that an authoritative DNS server
for gaia.cs.umass.edu is called dns.umass.edu. As shown in Figure
2.21, the host cis.poly.edu first sends a DNS query message to its local DNS
server, dns.poly.edu. The query message contains the hostname to be translated, namely, gaia.cs.umass.edu. The local DNS server forwards the query
message to a root DNS server. The root DNS server takes note of the edu suffix and
returns to the local DNS server a list of IP addresses for TLD servers responsible for
edu. The local DNS server then resends the query message to one of these TLD
servers. The TLD server takes note of the umass.edu suffix and responds with the
IP address of the authoritative DNS server for the University of Massachusetts,
namely, dns.umass.edu. Finally, the local DNS server resends the query
message directly to dns.umass.edu, which responds with the IP address of
gaia.cs.umass.edu. Note that in this example, in order to obtain the mapping
for one hostname, eight DNS messages were sent: four query messages and four
reply messages! We’ll soon see how DNS caching reduces this query traffic.
Our previous example assumed that the TLD server knows the authoritative
DNS server for the hostname. In general this not always true. Instead, the TLD server
may know only of an intermediate DNS server, which in turn knows the authoritative
DNS server for the hostname. For example, suppose again that the University of
2.5
•
DNS—THE INTERNET’S DIRECTORY SERVICE
Root DNS server
2
3
4
5
Local DNS server
dns.poly.edu
TLD DNS server
7
1
6
8
Authoritative DNS server
dns.umass.edu
Requesting host
cis.poly.edu
gaia.cs.umass.edu
Figure 2.21
Interaction of the various DNS servers
Massachusetts has a DNS server for the university, called dns.umass.edu. Also
suppose that each of the departments at the University of Massachusetts has its own
DNS server, and that each departmental DNS server is authoritative for all hosts in
the department. In this case, when the intermediate DNS server, dns.umass.edu,
receives a query for a host with a hostname ending with cs.umass.edu, it returns
to dns.poly.edu the IP address of dns.cs.umass.edu, which is authoritative for all hostnames ending with cs.umass.edu. The local DNS server
dns.poly.edu then sends the query to the authoritative DNS server, which
returns the desired mapping to the local DNS server, which in turn returns the mapping to the requesting host. In this case, a total of 10 DNS messages are sent!
The example shown in Figure 2.21 makes use of both recursive queries and
iterative queries. The query sent from cis.poly.edu to dns.poly.edu is a
recursive query, since the query asks dns.poly.edu to obtain the mapping on its
137
138
CHAPTER 2
•
APPLICATION LAYER
Root DNS server
2
7
6
Local DNS server
dns.poly.edu
3
TLD DNS server
5
1
4
8
Requesting host
cis.poly.edu
Authoritative DNS server
dns.umass.edu
gaia.cs.umass.edu
Figure 2.22
Recursive queries in DNS
behalf. But the subsequent three queries are iterative since all of the replies are
directly returned to dns.poly.edu. In theory, any DNS query can be iterative or
recursive. For example, Figure 2.22 shows a DNS query chain for which all of
the queries are recursive. In practice, the queries typically follow the pattern in
Figure 2.21: The query from the requesting host to the local DNS server is recursive, and the remaining queries are iterative.
DNS Caching
Our discussion thus far has ignored DNS caching, a critically important feature of the
DNS system. In truth, DNS extensively exploits DNS caching in order to improve
the delay performance and to reduce the number of DNS messages ricocheting around
2.5
•
DNS—THE INTERNET’S DIRECTORY SERVICE
the Internet. The idea behind DNS caching is very simple. In a query chain, when a
DNS server receives a DNS reply (containing, for example, a mapping from a hostname to an IP address), it can cache the mapping in its local memory. For example,
in Figure 2.21, each time the local DNS server dns.poly.edu receives a reply
from some DNS server, it can cache any of the information contained in the reply. If a
hostname/IP address pair is cached in a DNS server and another query arrives to the
DNS server for the same hostname, the DNS server can provide the desired IP address,
even if it is not authoritative for the hostname. Because hosts and mappings between
hostnames and IP addresses are by no means permanent, DNS servers discard cached
information after a period of time (often set to two days).
As an example, suppose that a host apricot.poly.edu queries
dns.poly.edu for the IP address for the hostname cnn.com. Furthermore, suppose that a few hours later, another Polytechnic University host, say, kiwi.poly.fr,
also queries dns.poly.edu with the same hostname. Because of caching, the local
DNS server will be able to immediately return the IP address of cnn.com to this second requesting host without having to query any other DNS servers. A local DNS
server can also cache the IP addresses of TLD servers, thereby allowing the local DNS
server to bypass the root DNS servers in a query chain (this often happens).
2.5.3 DNS Records and Messages
The DNS servers that together implement the DNS distributed database store
resource records (RRs), including RRs that provide hostname-to-IP address mappings. Each DNS reply message carries one or more resource records. In this and
the following subsection, we provide a brief overview of DNS resource records and
messages; more details can be found in [Abitz 1993] or in the DNS RFCs [RFC
1034; RFC 1035].
A resource record is a four-tuple that contains the following fields:
(Name, Value, Type, TTL)
TTL is the time to live of the resource record; it determines when a resource should
be removed from a cache. In the example records given below, we ignore the TTL
field. The meaning of Name and Value depend on Type:
• If Type=A, then Name is a hostname and Value is the IP address for the hostname. Thus, a Type A record provides the standard hostname-to-IP address mapping. As an example, (relay1.bar.foo.com, 145.37.93.126, A)
is a Type A record.
• If Type=NS, then Name is a domain (such as foo.com) and Value is the hostname of an authoritative DNS server that knows how to obtain the IP addresses
for hosts in the domain. This record is used to route DNS queries further along in
139
140
CHAPTER 2
•
APPLICATION LAYER
the query chain. As an example, (foo.com, dns.foo.com, NS) is a Type
NS record.
• If Type=CNAME, then Value is a canonical hostname for the alias hostname
Name. This record can provide querying hosts the canonical name for a hostname. As an example, (foo.com, relay1.bar.foo.com, CNAME) is a
CNAME record.
• If Type=MX, then Value is the canonical name of a mail server that has an alias
hostname Name. As an example, (foo.com, mail.bar.foo.com, MX)
is an MX record. MX records allow the hostnames of mail servers to have simple aliases. Note that by using the MX record, a company can have the same
aliased name for its mail server and for one of its other servers (such as its Web
server). To obtain the canonical name for the mail server, a DNS client would
query for an MX record; to obtain the canonical name for the other server, the
DNS client would query for the CNAME record.
If a DNS server is authoritative for a particular hostname, then the DNS server will
contain a Type A record for the hostname. (Even if the DNS server is not authoritative,
it may contain a Type A record in its cache.) If a server is not authoritative for a hostname, then the server will contain a Type NS record for the domain that includes the
hostname; it will also contain a Type A record that provides the IP address of the DNS
server in the Value field of the NS record. As an example, suppose an edu TLD server
is not authoritative for the host gaia.cs.umass.edu. Then this server will contain
a record for a domain that includes the host gaia.cs.umass.edu, for example,
(umass.edu, dns.umass.edu, NS). The edu TLD server would also contain
a Type A record, which maps the DNS server dns.umass.edu to an IP address, for
example, (dns.umass.edu, 128.119.40.111, A).
DNS Messages
Earlier in this section, we referred to DNS query and reply messages. These are the
only two kinds of DNS messages. Furthermore, both query and reply messages have
the same format, as shown in Figure 2.23.The semantics of the various fields in a
DNS message are as follows:
• The first 12 bytes is the header section, which has a number of fields. The first field
is a 16-bit number that identifies the query. This identifier is copied into the reply
message to a query, allowing the client to match received replies with sent queries.
There are a number of flags in the flag field. A 1-bit query/reply flag indicates
whether the message is a query (0) or a reply (1). A 1-bit authoritative flag is set in a
reply message when a DNS server is an authoritative server for a queried name. A
1-bit recursion-desired flag is set when a client (host or DNS server) desires that the
DNS server perform recursion when it doesn’t have the record. A 1-bit recursionavailable field is set in a reply if the DNS server supports recursion. In the header,
2.5
Flags
Number of questions
Number of answer RRs
Number of authority RRs
Number of additional RRs
Name, type fields for
a query
RRs in response to query
Authority
(variable number of resource records)
Records for
authoritative servers
Figure 2.23
•
•
12 bytes
Answers
(variable number of resource records)
Additional information
(variable number of resource records)
•
DNS—THE INTERNET’S DIRECTORY SERVICE
Identification
Questions
(variable number of questions)
•
•
Additional “helpful”
info that may be used
DNS message format
there are also four number-of fields. These fields indicate the number of occurrences
of the four types of data sections that follow the header.
The question section contains information about the query that is being made.
This section includes (1) a name field that contains the name that is being
queried, and (2) a type field that indicates the type of question being asked about
the name—for example, a host address associated with a name (Type A) or the
mail server for a name (Type MX).
In a reply from a DNS server, the answer section contains the resource records
for the name that was originally queried. Recall that in each resource record there
is the Type (for example, A, NS, CNAME, and MX), the Value, and the TTL.
A reply can return multiple RRs in the answer, since a hostname can have multiple IP addresses (for example, for replicated Web servers, as discussed earlier in
this section).
The authority section contains records of other authoritative servers.
The additional section contains other helpful records. For example, the answer
field in a reply to an MX query contains a resource record providing the canonical hostname of a mail server. The additional section contains a Type A record
providing the IP address for the canonical hostname of the mail server.
How would you like to send a DNS query message directly from the host
you’re working on to some DNS server? This can easily be done with the nslookup
141
142
CHAPTER 2
•
APPLICATION LAYER
program, which is available from most Windows and UNIX platforms. For example, from a Windows host, open the Command Prompt and invoke the nslookup program by simply typing “nslookup.” After invoking nslookup, you can send a DNS
query to any DNS server (root, TLD, or authoritative). After receiving the reply
message from the DNS server, nslookup will display the records included in the
reply (in a human-readable format). As an alternative to running nslookup from your
own host, you can visit one of many Web sites that allow you to remotely employ
nslookup. (Just type “nslookup” into a search engine and you’ll be brought to one of
these sites.) The DNS Wireshark lab at the end of this chapter will allow you to
explore the DNS in much more detail.
Inserting Records into the DNS Database
The discussion above focused on how records are retrieved from the DNS database.
You might be wondering how records get into the database in the first place. Let’s look
at how this is done in the context of a specific example. Suppose you have just created
an exciting new startup company called Network Utopia. The first thing you’ll surely
want to do is register the domain name networkutopia.com at a registrar. A
registrar is a commercial entity that verifies the uniqueness of the domain name,
enters the domain name into the DNS database (as discussed below), and collects a
small fee from you for its services. Prior to 1999, a single registrar, Network Solutions,
had a monopoly on domain name registration for com, net, and org domains. But
now there are many registrars competing for customers, and the Internet Corporation
for Assigned Names and Numbers (ICANN) accredits the various registrars. A complete list of accredited registrars is available at http://www.internic.net.
When you register the domain name networkutopia.com with some registrar, you also need to provide the registrar with the names and IP addresses of your
primary and secondary authoritative DNS servers. Suppose the names and IP
addresses are dns1.networkutopia.com, dns2.networkutopia.com,
212.212.212.1, and 212.212.212.2. For each of these two authoritative
DNS servers, the registrar would then make sure that a Type NS and a Type A record
are entered into the TLD com servers. Specifically, for the primary authoritative
server for networkutopia.com, the registrar would insert the following two
resource records into the DNS system:
(networkutopia.com, dns1.networkutopia.com, NS)
(dns1.networkutopia.com, 212.212.212.1, A)
You’ll also have to make sure that the Type A resource record for your Web server
www.networkutopia.com and the Type MX resource record for your mail
server mail.networkutopia.com are entered into your authoritative DNS
servers. (Until recently, the contents of each DNS server were configured statically,
2.5
•
DNS—THE INTERNET’S DIRECTORY SERVICE
FOCUS ON SECURITY
DNS VULNERABILITIES
We have seen that DNS is a critical component of the Internet infrastructure, with
many important services - including the Web and e-mail - simply incapable of functioning without it. We therefore naturally ask, how can DNS be attacked? Is DNS a
sitting duck, waiting to be knocked out of service, while taking most Internet applications down with it?
The first type of attack that comes to mind is a DDoS bandwidth-flooding attack (see
Section 1.6) against DNS servers. For example, an attacker could attempt to send to
each DNS root server a deluge of packets, so many that the majority of legitimate DNS
queries never get answered. Such a large-scale DDoS attack against DNS root servers
actually took place on October 21, 2002. In this attack, the attackers leveraged a botnet to send truck loads of ICMP ping messages to each of the 13 DNS root servers.
(ICMP messages are discussed in Chapter 4. For now, it suffices to know that ICMP packets are special types of IP datagrams.) Fortunately, this large-scale attack caused minimal
damage, having little or no impact on users’ Internet experience. The attackers did
succeed at directing a deluge of packets at the root servers. But many of the DNS root
servers were protected by packet filters, configured to always block all ICMP ping
messages directed at the root servers. These protected servers were thus spared and
functioned as normal. Furthermore, most local DNS servers cache the IP addresses of toplevel-domain servers, allowing the query process to often bypass the DNS root servers.
A potentially more effective DDoS attack against DNS would be send a deluge of
DNS queries to top-level-domain servers, for example, to all the top-level-domain
servers that handle the .com domain. It would be harder to filter DNS queries directed to DNS servers; and top-level-domain servers are not as easily bypassed as are
root servers. But the severity of such an attack would be partially mitigated by
caching in local DNS servers.
DNS could potentially be attacked in other ways. In a man-in-the-middle attack,
the attacker intercepts queries from hosts and returns bogus replies. In the DNS poisoning attack, the attacker sends bogus replies to a DNS server, tricking the server
into accepting bogus records into its cache. Either of these attacks could be used, for
example, to redirect an unsuspecting Web user to the attacker’s Web site. These
attacks, however, are difficult to implement, as they require intercepting packets or
throttling servers [Skoudis 2006].
Another important DNS attack is not an attack on the DNS service per se, but
instead exploits the DNS infrastructure to launch a DDoS attack against a targeted host
(for example, your university’s mail server). In this attack, the attacker sends DNS
queries to many authoritative DNS servers, with each query having the spoofed source
address of the targeted host. The DNS servers then send their replies directly to the targeted host. If the queries can be crafted in such a way that a response is much larger
143
144
CHAPTER 2
•
APPLICATION LAYER
FOCUS ON SECURITY
(in bytes) than a query (so-called amplification), then the attacker can potentially overwhelm the target without having to generate much of its own traffic. Such reflection
attacks exploiting DNS have had limited success to date [Mirkovic 2005].
In summary, DNS has demonstrated itself to be surprisingly robust against attacks.
To date, there hasn’t been an attack that has successfully impeded the DNS service.
There have been successful reflector attacks; however, these attacks can be (and are
being) addressed by appropriate configuration of DNS servers.
for example, from a configuration file created by a system manager. More recently,
an UPDATE option has been added to the DNS protocol to allow data to be dynamically added or deleted from the database via DNS messages. [RFC 2136] and [RFC
3007] specify DNS dynamic updates.)
Once all of these steps are completed, people will be able to visit your Web site
and send e-mail to the employees at your company. Let’s conclude our discussion of
DNS by verifying that this statement is true. This verification also helps to solidify
what we have learned about DNS. Suppose Alice in Australia wants to view the Web
page www.networkutopia.com. As discussed earlier, her host will first send a
DNS query to her local DNS server. The local DNS server will then contact a TLD
com server. (The local DNS server will also have to contact a root DNS server if the
address of a TLD com server is not cached.) This TLD server contains the Type NS
and Type A resource records listed above, because the registrar had these resource
records inserted into all of the TLD com servers. The TLD com server sends a reply
to Alice’s local DNS server, with the reply containing the two resource records. The
local DNS server then sends a DNS query to 212.212.212.1, asking for the
Type A record corresponding to www.networkutopia.com. This record provides the IP address of the desired Web server, say, 212.212.71.4, which the
local DNS server passes back to Alice’s host. Alice’s browser can now initiate a TCP
connection to the host 212.212.71.4 and send an HTTP request over the connection. Whew! There’s a lot more going on than what meets the eye when one surfs
the Web!
2.6 Peer-to-Peer Applications
The applications described in this chapter thus far—including the Web, e-mail, and
DNS—all employ client-server architectures with significant reliance on always-on
infrastructure servers. Recall from Section 2.1.1 that with a P2P architecture, there
is minimal (or no) reliance on always-on infrastructure servers. Instead, pairs of
intermittently connected hosts, called peers, communicate directly with each other.
2.6
•
PEER-TO-PEER APPLICATIONS
The peers are not owned by a service provider, but are instead desktops and laptops
controlled by users.
In this section we’ll examine two different applications that are particularly
well-suited for P2P designs. The first is file distribution, where the application distributes a file from a single source to a large number of peers. File distribution is a
nice place to start our investigation of P2P, as it clearly exposes the self-scalability
of P2P architectures. As a specific example for file distribution, we’ll describe
the popular BitTorrent system. The second P2P application we’ll examine is a
database distributed over a large community of peers. For this application, we’ll
explore the concept of a Distributed Hash Table (DHT).
2.6.1 P2P File Distribution
We begin our foray into P2P by considering a very natural application, namely,
distributing a large file from a single server to a large number of hosts (called
peers). The file might be a new version of the Linux operating system, a software
patch for an existing operating system or application, an MP3 music file, or an
MPEG video file. In client-server file distribution, the server must send a copy of
the file to each of the peers—placing an enormous burden on the server and consuming a large amount of server bandwidth. In P2P file distribution, each peer can
redistribute any portion of the file it has received to any other peers, thereby
assisting the server in the distribution process. As of 2012, the most popular P2P
file distribution protocol is BitTorrent. Originally developed by Bram Cohen,
there are now many different independent BitTorrent clients conforming to the
BitTorrent protocol, just as there are a number of Web browser clients that
conform to the HTTP protocol. In this subsection, we first examine the selfscalability of P2P architectures in the context of file distribution. We then describe
BitTorrent in some detail, highlighting its most important characteristics and
features.
Scalability of P2P Architectures
To compare client-server architectures with peer-to-peer architectures, and illustrate
the inherent self-scalability of P2P, we now consider a simple quantitative model for
distributing a file to a fixed set of peers for both architecture types. As shown in Figure 2.24, the server and the peers are connected to the Internet with access links.
Denote the upload rate of the server’s access link by us, the upload rate of the ith
peer’s access link by ui, and the download rate of the ith peer’s access link by di.
Also denote the size of the file to be distributed (in bits) by F and the number of
peers that want to obtain a copy of the file by N. The distribution time is the time it
takes to get a copy of the file to all N peers. In our analysis of the distribution time
below, for both client-server and P2P architectures, we make the simplifying (and
generally accurate [Akella 2003]) assumption that the Internet core has abundant
145
146
CHAPTER 2
•
APPLICATION LAYER
File: F
Server
u1
d1
u2
us
d2
u3
d3
dN
Internet
uN
u4
d4
d6
u6
Figure 2.24
d5
u5
An illustrative file distribution problem
bandwidth, implying that all of the bottlenecks are in access networks. We also suppose that the server and clients are not participating in any other network applications, so that all of their upload and download access bandwidth can be fully
devoted to distributing this file.
Let’s first determine the distribution time for the client-server architecture,
which we denote by Dcs. In the client-server architecture, none of the peers aids in
distributing the file. We make the following observations:
• The server must transmit one copy of the file to each of the N peers. Thus the
server must transmit NF bits. Since the server’s upload rate is us, the time to distribute the file must be at least NF/us.
• Let dmin denote the download rate of the peer with the lowest download rate, that
is, dmin = min{d1,dp,...,dN}. The peer with the lowest download rate cannot
obtain all F bits of the file in less than F/dmin seconds. Thus the minimum distribution time is at least F/dmin.
Putting these two observations together, we obtain
Dcs Ú max b
NF F
,
r.
u s dmin
2.6
•
PEER-TO-PEER APPLICATIONS
This provides a lower bound on the minimum distribution time for the client-server
architecture. In the homework problems you will be asked to show that the server
can schedule its transmissions so that the lower bound is actually achieved. So let’s
take this lower bound provided above as the actual distribution time, that is,
Dcs = max b
NF F
,
r
u s dmin
(2.1)
We see from Equation 2.1 that for N large enough, the client-server distribution time
is given by NF/us. Thus, the distribution time increases linearly with the number of
peers N. So, for example, if the number of peers from one week to the next increases
a thousand-fold from a thousand to a million, the time required to distribute the file
to all peers increases by 1,000.
Let’s now go through a similar analysis for the P2P architecture, where each
peer can assist the server in distributing the file. In particular, when a peer receives
some file data, it can use its own upload capacity to redistribute the data to other
peers. Calculating the distribution time for the P2P architecture is somewhat more
complicated than for the client-server architecture, since the distribution time
depends on how each peer distributes portions of the file to the other peers. Nevertheless, a simple expression for the minimal distribution time can be obtained
[Kumar 2006]. To this end, we first make the following observations:
• At the beginning of the distribution, only the server has the file. To get this file
into the community of peers, the server must send each bit of the file at least once
into its access link. Thus, the minimum distribution time is at least F/us. (Unlike
the client-server scheme, a bit sent once by the server may not have to be sent by
the server again, as the peers may redistribute the bit among themselves.)
• As with the client-server architecture, the peer with the lowest download rate
cannot obtain all F bits of the file in less than F/dmin seconds. Thus the minimum
distribution time is at least F/dmin.
• Finally, observe that the total upload capacity of the system as a whole is equal
to the upload rate of the server plus the upload rates of each of the individual
peers, that is, utotal = us + u1 + … + uN. The system must deliver (upload) F bits
to each of the N peers, thus delivering a total of NF bits. This cannot be done at a
rate faster than u total . Thus, the minimum distribution time is also at least
NF/(us + u1 + … + uN).
Putting these three observations together, we obtain the minimum distribution time
for P2P, denoted by DP2P.
F F
,
,
DP2P Ú max c u s dmin
NF
N
us + a ui
i=1
s
(2.2)
147
CHAPTER 2
•
APPLICATION LAYER
Equation 2.2 provides a lower bound for the minimum distribution time for the P2P
architecture. It turns out that if we imagine that each peer can redistribute a bit as
soon as it receives the bit, then there is a redistribution scheme that actually achieves
this lower bound [Kumar 2006]. (We will prove a special case of this result in the
homework.) In reality, where chunks of the file are redistributed rather than individual bits, Equation 2.2 serves as a good approximation of the actual minimum distribution time. Thus, let’s take the lower bound provided by Equation 2.2 as the actual
minimum distribution time, that is,
F F
,
,
DP2P = max c u s dmin
NF
N
us + a ui
s
(2.3)
i=1
Figure 2.25 compares the minimum distribution time for the client-server and
P2P architectures assuming that all peers have the same upload rate u. In Figure
2.25, we have set F/u = 1 hour, us = 10u, and dmin ≥ us. Thus, a peer can transmit the
entire file in one hour, the server transmission rate is 10 times the peer upload rate,
and (for simplicity) the peer download rates are set large enough so as not to have
an effect. We see from Figure 2.25 that for the client-server architecture, the distribution time increases linearly and without bound as the number of peers
increases. However, for the P2P architecture, the minimal distribution time is not
only always less than the distribution time of the client-server architecture; it is also
less than one hour for any number of peers N. Thus, applications with the P2P
architecture can be self-scaling. This scalability is a direct consequence of peers
being redistributors as well as consumers of bits.
3.5
3.0
Minimum distributioin tiime
148
Client-Server
2.5
2.0
1.5
P2P
1.0
0.5
0
0
5
10
15
20
25
30
35
N
Figure 2.25
Distribution time for P2P and client-server architectures
2.6
•
PEER-TO-PEER APPLICATIONS
BitTorrent
BitTorrent is a popular P2P protocol for file distribution [Chao 2011]. In BitTorrent lingo, the collection of all peers participating in the distribution of a particular
file is called a torrent. Peers in a torrent download equal-size chunks of the file
from one another, with a typical chunk size of 256 KBytes. When a peer first joins
a torrent, it has no chunks. Over time it accumulates more and more chunks. While
it downloads chunks it also uploads chunks to other peers. Once a peer has
acquired the entire file, it may (selfishly) leave the torrent, or (altruistically) remain
in the torrent and continue to upload chunks to other peers. Also, any peer may leave
the torrent at any time with only a subset of chunks, and later rejoin the torrent.
Let’s now take a closer look at how BitTorrent operates. Since BitTorrent is a
rather complicated protocol and system, we’ll only describe its most important
mechanisms, sweeping some of the details under the rug; this will allow us to see
the forest through the trees. Each torrent has an infrastructure node called a tracker.
When a peer joins a torrent, it registers itself with the tracker and periodically
informs the tracker that it is still in the torrent. In this manner, the tracker keeps
track of the peers that are participating in the torrent. A given torrent may have
fewer than ten or more than a thousand peers participating at any instant of time.
As shown in Figure 2.26, when a new peer, Alice, joins the torrent, the tracker
randomly selects a subset of peers (for concreteness, say 50) from the set of participating peers, and sends the IP addresses of these 50 peers to Alice. Possessing this list of
peers, Alice attempts to establish concurrent TCP connections with all the peers on this
list. Let’s call all the peers with which Alice succeeds in establishing a TCP connection “neighboring peers.” (In Figure 2.26, Alice is shown to have only three neighboring peers. Normally, she would have many more.) As time evolves, some of these
peers may leave and other peers (outside the initial 50) may attempt to establish TCP
connections with Alice. So a peer’s neighboring peers will fluctuate over time.
At any given time, each peer will have a subset of chunks from the file, with different peers having different subsets. Periodically, Alice will ask each of her neighboring peers (over the TCP connections) for the list of the chunks they have. If Alice has L
different neighbors, she will obtain L lists of chunks. With this knowledge, Alice will
issue requests (again over the TCP connections) for chunks she currently does not have.
So at any given instant of time, Alice will have a subset of chunks and will
know which chunks her neighbors have. With this information, Alice will have two
important decisions to make. First, which chunks should she request first from her
neighbors? And second, to which of her neighbors should she send requested
chunks? In deciding which chunks to request, Alice uses a technique called rarest
first. The idea is to determine, from among the chunks she does not have, the
chunks that are the rarest among her neighbors (that is, the chunks that have the
fewest repeated copies among her neighbors) and then request those rarest chunks
first. In this manner, the rarest chunks get more quickly redistributed, aiming to
(roughly) equalize the numbers of copies of each chunk in the torrent.
149
150
CHAPTER 2
•
APPLICATION LAYER
Tracker
Peer
Obtain
list of
peers
Trading chunks
Alice
Figure 2.26
File distribution with BitTorrent
To determine which requests she responds to, BitTorrent uses a clever trading
algorithm. The basic idea is that Alice gives priority to the neighbors that are currently supplying her data at the highest rate. Specifically, for each of her neighbors,
Alice continually measures the rate at which she receives bits and determines the four
peers that are feeding her bits at the highest rate. She then reciprocates by sending
chunks to these same four peers. Every 10 seconds, she recalculates the rates and possibly modifies the set of four peers. In BitTorrent lingo, these four peers are said to
be unchoked. Importantly, every 30 seconds, she also picks one additional neighbor
at random and sends it chunks. Let’s call the randomly chosen peer Bob. In BitTorrent lingo, Bob is said to be optimistically unchoked. Because Alice is sending data
to Bob, she may become one of Bob’s top four uploaders, in which case Bob would
start to send data to Alice. If the rate at which Bob sends data to Alice is high enough,
Bob could then, in turn, become one of Alice’s top four uploaders. In other words,
every 30 seconds, Alice will randomly choose a new trading partner and initiate trading with that partner. If the two peers are satisfied with the trading, they will put each
other in their top four lists and continue trading with each other until one of the peers
finds a better partner. The effect is that peers capable of uploading at compatible rates
tend to find each other. The random neighbor selection also allows new peers to get
2.6
•
PEER-TO-PEER APPLICATIONS
chunks, so that they can have something to trade. All other neighboring peers besides
these five peers (four “top” peers and one probing peer) are “choked,” that is, they do
not receive any chunks from Alice. BitTorrent has a number of interesting mechanisms that are not discussed here, including pieces (mini-chunks), pipelining, random
first selection, endgame mode, and anti-snubbing [Cohen 2003].
The incentive mechanism for trading just described is often referred to as tit-for-tat
[Cohen 2003]. It has been shown that this incentive scheme can be circumvented
[Liogkas 2006; Locher 2006; Piatek 2007]. Nevertheless, the BitTorrent ecosystem is
wildly successful, with millions of simultaneous peers actively sharing files in hundreds of thousands of torrents. If BitTorrent had been designed without tit-for-tat (or a
variant), but otherwise exactly the same, BitTorrent would likely not even exist now, as
the majority of the users would have been freeriders [Saroiu 2002].
Interesting variants of the BitTorrent protocol are proposed [Guo 2005; Piatek
2007]. Also, many of the P2P live streaming applications, such as PPLive and
ppstream, have been inspired by BitTorrent [Hei 2007].
2.6.2 Distributed Hash Tables (DHTs)
In this section, we will consider how to implement a simple database in a P2P network. Let’s begin by describing a centralized version of this simple database, which
will simply contain (key, value) pairs. For example, the keys could be social security numbers and the values could be the corresponding human names; in this case,
an example key-value pair is (156-45-7081, Johnny Wu). Or the keys could be content names (e.g., names of movies, albums, and software), and the value could be
the IP address at which the content is stored; in this case, an example key-value pair
is (Led Zeppelin IV, 128.17.123.38). We query the database with a key. If there are
one or more key-value pairs in the database that match the query key, the database
returns the corresponding values. So, for example, if the database stores social security numbers and their corresponding human names, we can query with a specific
social security number, and the database returns the name of the human who has that
social security number. Or, if the database stores content names and their corresponding IP addresses, we can query with a specific content name, and the database
returns the IP addresses that store the specific content.
Building such a database is straightforward with a client-server architecture that
stores all the (key, value) pairs in one central server. So in this section, we’ll instead
consider how to build a distributed, P2P version of this database that will store the
(key, value) pairs over millions of peers. In the P2P system, each peer will only hold a
small subset of the totality of the (key, value) pairs. We’ll allow any peer to query the
distributed database with a particular key. The distributed database will then locate the
peers that have the corresponding (key, value) pairs and return the key-value pairs to
the querying peer. Any peer will also be allowed to insert new key-value pairs into the
database. Such a distributed database is referred to as a distributed hash table
(DHT).
151
152
CHAPTER 2
VideoNote
Walking through
distributed hash tables
•
APPLICATION LAYER
Before describing how we can create a DHT, let’s first describe a specific
example DHT service in the context of P2P file sharing. In this case, a key is the
content name and the value is the IP address of a peer that has a copy of the content.
So, if Bob and Charlie each have a copy of the latest Linux distribution, then the
DHT database will include the following two key-value pairs: (Linux, IPBob) and
(Linux, IPCharlie). More specifically, since the DHT database is distributed over the
peers, some peer, say Dave, will be responsible for the key “Linux” and will have
the corresponding key-value pairs. Now suppose Alice wants to obtain a copy of
Linux. Clearly, she first needs to know which peers have a copy of Linux before she
can begin to download it. To this end, she queries the DHT with “Linux” as the key.
The DHT then determines that the peer Dave is responsible for the key “Linux.” The
DHT then contacts peer Dave, obtains from Dave the key-value pairs (Linux, IPBob)
and (Linux, IPCharlie), and passes them on to Alice. Alice can then download the latest Linux distribution from either IPBob or IPCharlie.
Now let’s return to the general problem of designing a DHT for general keyvalue pairs. One naïve approach to building a DHT is to randomly scatter the (key,
value) pairs across all the peers and have each peer maintain a list of the IP
addresses of all participating peers. In this design, the querying peer sends its query
to all other peers, and the peers containing the (key, value) pairs that match the key
can respond with their matching pairs. Such an approach is completely unscalable,
of course, as it would require each peer to not only know about all other peers (possibly millions of such peers!) but even worse, have each query sent to all peers.
We now describe an elegant approach to designing a DHT. To this end, let’s first
assign an identifier to each peer, where each identifier is an integer in the range [0,
2n ⫺ 1] for some fixed n. Note that each such identifier can be expressed by an n-bit
representation. Let’s also require each key to be an integer in the same range. The
astute reader may have observed that the example keys described a little earlier (social
security numbers and content names) are not integers. To create integers out of such
keys, we will use a hash function that maps each key (e.g., social security number) to
an integer in the range [0, 2n ⫺ 1]. A hash function is a many-to-one function for which
two different inputs can have the same output (same integer), but the likelihood of the
having the same output is extremely small. (Readers who are unfamiliar with hash
functions may want to visit Chapter 7, in which hash functions are discussed in some
detail.) The hash function is assumed to be available to all peers in the system. Henceforth, when we refer to the “key,” we are referring to the hash of the original key. So,
for example, if the original key is “Led Zeppelin IV,” the key used in the DHT will be
the integer that equals the hash of “Led Zeppelin IV.” As you may have guessed, this
is why “Hash” is used in the term “Distributed Hash Function.”
Let’s now consider the problem of storing the (key, value) pairs in the DHT. The
central issue here is defining a rule for assigning keys to peers. Given that each peer
has an integer identifier and that each key is also an integer in the same range, a natural approach is to assign each (key, value) pair to the peer whose identifier is the
closest to the key. To implement such a scheme, we’ll need to define what is meant by
“closest,” for which many conventions are possible. For convenience, let’s define the
2.6
•
PEER-TO-PEER APPLICATIONS
closest peer as the closest successor of the key. To gain some insight here, let’s take a
look at a specific example. Suppose n ⫽ 4 so that all the peer and key identifiers are in
the range [0, 15]. Further suppose that there are eight peers in the system with identifiers 1, 3, 4, 5, 8, 10, 12, and 15. Finally, suppose we want to store the (key, value) pair
(11, Johnny Wu) in one of the eight peers. But in which peer? Using our closest convention, since peer 12 is the closest successor for key 11, we therefore store the pair
(11, Johnny Wu) in the peer 12. [To complete our definition of closest, if the key is
exactly equal to one of the peer identifiers, we store the (key, value) pair in that matching peer; and if the key is larger than all the peer identifiers, we use a modulo-2n convention, storing the (key, value) pair in the peer with the smallest identifier.]
Now suppose a peer, Alice, wants to insert a (key, value) pair into the DHT.
Conceptually, this is straightforward: She first determines the peer whose identifier
is closest to the key; she then sends a message to that peer, instructing it to store the
(key, value) pair. But how does Alice determine the peer that is closest to the key? If
Alice were to keep track of all the peers in the system (peer IDs and corresponding
IP addresses), she could locally determine the closest peer. But such an approach
requires each peer to keep track of all other peers in the DHT—which is completely
impractical for a large-scale system with millions of peers.
Circular DHT
To address this problem of scale, let’s now consider organizing the peers into a
circle. In this circular arrangement, each peer only keeps track of its immediate successor and immediate predecessor (modulo 2n). An example of such a circle is
shown in Figure 2.27(a). In this example, n is again 4 and there are the same eight
1
Who is
responsible
for key 11?
1
3
3
15
15
4
12
4
12
5
5
10
10
8
a.
Figure 2.27
8
b.
(a) A circular DHT. Peer 3 wants to determine who is
responsible for key 11. (b) A circular DHT with shortcuts
153
154
CHAPTER 2
•
APPLICATION LAYER
peers from the previous example. Each peer is only aware of its immediate successor and predecessor; for example, peer 5 knows the IP address and identifier for
peers 8 and 4 but does not necessarily know anything about any other peers that may
be in the DHT. This circular arrangement of the peers is a special case of an overlay
network. In an overlay network, the peers form an abstract logical network which
resides above the “underlay” computer network consisting of physical links, routers,
and hosts. The links in an overlay network are not physical links, but are simply virtual liaisons between pairs of peers. In the overlay in Figure 2.27(a), there are eight
peers and eight overlay links; in the overlay in Figure 2.27(b) there are eight peers
and 16 overlay links. A single overlay link typically uses many physical links and
physical routers in the underlay network.
Using the circular overlay in Figure 2.27(a), now suppose that peer 3 wants to
determine which peer in the DHT is responsible for key 11. Using the circular overlay,
the origin peer (peer 3) creates a message saying “Who is responsible for key 11?” and
sends this message clockwise around the circle. Whenever a peer receives such a message, because it knows the identifier of its successor and predecessor, it can determine
whether it is responsible for (that is, closest to) the key in question. If a peer is not
responsible for the key, it simply sends the message to its successor. So, for example,
when peer 4 receives the message asking about key 11, it determines that it is not
responsible for the key (because its successor is closer to the key), so it just passes the
message along to peer 5. This process continues until the message arrives at peer 12,
who determines that it is the closest peer to key 11. At this point, peer 12 can send a
message back to the querying peer, peer 3, indicating that it is responsible for key 11.
The circular DHT provides a very elegant solution for reducing the amount of
overlay information each peer must manage. In particular, each peer needs only to
be aware of two peers, its immediate successor and its immediate predecessor. But
this solution introduces yet a new problem. Although each peer is only aware of two
neighboring peers, to find the node responsible for a key (in the worst case), all N
nodes in the DHT will have to forward a message around the circle; N/2 messages
are sent on average.
Thus, in designing a DHT, there is tradeoff between the number of neighbors each
peer has to track and the number of messages that the DHT needs to send to resolve a
single query. On one hand, if each peer tracks all other peers (mesh overlay), then only
one message is sent per query, but each peer has to keep track of N peers. On the other
hand, with a circular DHT, each peer is only aware of two peers, but N/2 messages are
sent on average for each query. Fortunately, we can refine our designs of DHTs so that
the number of neighbors per peer as well as the number of messages per query is kept
to an acceptable size. One such refinement is to use the circular overlay as a foundation, but add “shortcuts” so that each peer not only keeps track of its immediate successor and predecessor, but also of a relatively small number of shortcut peers
scattered about the circle. An example of such a circular DHT with some shortcuts is
shown in Figure 2.27(b). Shortcuts are used to expedite the routing of query messages.
Specifically, when a peer receives a message that is querying for a key, it forwards the
2.6
•
PEER-TO-PEER APPLICATIONS
message to the neighbor (successor neighbor or one of the shortcut neighbors) which
is the closet to the key. Thus, in Figure 2.27(b), when peer 4 receives the message asking about key 11, it determines that the closet peer to the key (among its neighbors) is
its shortcut neighbor 10 and then forwards the message directly to peer 10. Clearly,
shortcuts can significantly reduce the number of messages used to process a query.
The next natural question is “How many shortcut neighbors should a peer have,
and which peers should be these shortcut neighbors? This question has received significant attention in the research community [Balakrishnan 2003; AndroutsellisTheotokis 2004]. Importantly, it has been shown that the DHT can be designed so that
both the number of neighbors per peer as well as the number of messages per query is
O(log N), where N is the number of peers. Such designs strike a satisfactory compromise between the extreme solutions of using mesh and circular overlay topologies.
Peer Churn
In P2P systems, a peer can come or go without warning. Thus, when designing a
DHT, we also must be concerned about maintaining the DHT overlay in the presence of such peer churn. To get a big-picture understanding of how this could be
accomplished, let’s once again consider the circular DHT in Figure 2.27(a). To handle peer churn, we will now require each peer to track (that is, know the IP address
of) its first and second successors; for example, peer 4 now tracks both peer 5 and
peer 8. We also require each peer to periodically verify that its two successors are
alive (for example, by periodically sending ping messages to them and asking for
responses). Let’s now consider how the DHT is maintained when a peer abruptly
leaves. For example, suppose peer 5 in Figure 2.27(a) abruptly leaves. In this case,
the two peers preceding the departed peer (4 and 3) learn that 5 has departed, since
it no longer responds to ping messages. Peers 4 and 3 thus need to update their successor state information. Let’s consider how peer 4 updates its state:
1. Peer 4 replaces its first successor (peer 5) with its second successor (peer 8).
2. Peer 4 then asks its new first successor (peer 8) for the identifier and IP address of
its immediate successor (peer 10). Peer 4 then makes peer 10 its second successor.
In the homework problems, you will be asked to determine how peer 3 updates its
overlay routing information.
Having briefly addressed what has to be done when a peer leaves, let’s now
consider what happens when a peer wants to join the DHT. Let’s say a peer with
identifier 13 wants to join the DHT, and at the time of joining, it only knows about
peer 1’s existence in the DHT. Peer 13 would first send peer 1 a message, saying
“what will be 13’s predecessor and successor?” This message gets forwarded
through the DHT until it reaches peer 12, who realizes that it will be 13’s predecessor and that its current successor, peer 15, will become 13’s successor. Next, peer 12
sends this predecessor and successor information to peer 13. Peer 13 can now join
155
156
CHAPTER 2
•
APPLICATION LAYER
the DHT by making peer 15 its successor and by notifying peer 12 that it should
change its immediate successor to 13.
DHTs have been finding widespread use in practice. For example, BitTorrent
uses the Kademlia DHT to create a distributed tracker. In the BitTorrent, the key is
the torrent identifier and the value is the IP addresses of all the peers currently participating in the torrent [Falkner 2007, Neglia 2007]. In this manner, by querying
the DHT with a torrent identifier, a newly arriving BitTorrent peer can determine the
peer that is responsible for the identifier (that is, for tracking the peers in the torrent). After having found that peer, the arriving peer can query it for a list of other
peers in the torrent.
2.7 Socket Programming: Creating Network
Applications
Now that we’ve looked at a number of important network applications, let’s explore
how network application programs are actually created. Recall from Section 2.1 that
a typical network application consists of a pair of programs—a client program and a
server program—residing in two different end systems. When these two programs
are executed, a client process and a server process are created, and these processes
communicate with each other by reading from, and writing to, sockets. When creating a network application, the developer’s main task is therefore to write the code
for both the client and server programs.
There are two types of network applications. One type is an implementation
whose operation is specified in a protocol standard, such as an RFC or some
other standards document; such an application is sometimes referred to as
“open,” since the rules specifying its operation are known to all. For such an
implementation, the client and server programs must conform to the rules dictated by the RFC. For example, the client program could be an implementation
of the client side of the FTP protocol, described in Section 2.3 and explicitly
defined in RFC 959; similarly, the server program could be an implementation of
the FTP server protocol, also explicitly defined in RFC 959. If one developer
writes code for the client program and another developer writes code for the
server program, and both developers carefully follow the rules of the RFC, then
the two programs will be able to interoperate. Indeed, many of today’s network
applications involve communication between client and server programs that
have been created by independent developers—for example, a Firefox browser
communicating with an Apache Web server, or a BitTorrent client communicating with BitTorrent tracker.
The other type of network application is a proprietary network application.
In this case the client and server programs employ an application-layer protocol that
has not been openly published in an RFC or elsewhere. A single developer (or
2.7
•
SOCKET PROGRAMMING: CREATING NETWORK APPLICATIONS
development team) creates both the client and server programs, and the developer
has complete control over what goes in the code. But because the code does not
implement an open protocol, other independent developers will not be able to
develop code that interoperates with the application.
In this section, we’ll examine the key issues in developing a client-server application, and we’ll “get our hands dirty” by looking at code that implements a very
simple client-server application. During the development phase, one of the first
decisions the developer must make is whether the application is to run over TCP
or over UDP. Recall that TCP is connection oriented and provides a reliable bytestream channel through which data flows between two end systems. UDP is
connectionless and sends independent packets of data from one end system to the
other, without any guarantees about delivery. Recall also that when a client or server
program implements a protocol defined by an RFC, it should use the well-known
port number associated with the protocol; conversely, when developing a proprietary application, the developer must be careful to avoid using such well-known
port numbers. (Port numbers were briefly discussed in Section 2.1. They are covered in more detail in Chapter 3.)
We introduce UDP and TCP socket programming by way of a simple UDP application and a simple TCP application. We present the simple UDP and TCP applications in Python. We could have written the code in Java, C, or C++, but we chose
Python mostly because Python clearly exposes the key socket concepts. With Python
there are fewer lines of code, and each line can be explained to the novice programmer without difficulty. But there’s no need to be frightened if you are not familiar with
Python. You should be able to easily follow the code if you have experience programming in Java, C, or C++.
If you are interested in client-server programming with Java, you are encouraged to see the companion Web site for this textbook; in fact, you can find there all
the examples in this section (and associated labs) in Java. For readers who are interested in client-server programming in C, there are several good references available
[Donahoo 2001; Stevens 1997; Frost 1994; Kurose 1996]; our Python examples
below have a similar look and feel to C.
2.7.1 Socket Programming with UDP
In this subsection, we’ll write simple client-server programs that use UDP; in the
following section, we’ll write similar programs that use TCP.
Recall from Section 2.1 that processes running on different machines communicate with each other by sending messages into sockets. We said that each process is
analogous to a house and the process’s socket is analogous to a door. The application
resides on one side of the door in the house; the transport-layer protocol resides on
the other side of the door in the outside world. The application developer has control
of everything on the application-layer side of the socket; however, it has little control
of the transport-layer side.
157
158
CHAPTER 2
•
APPLICATION LAYER
Now let’s take a closer look at the interaction between two communicating
processes that use UDP sockets. Before the sending process can push a packet of
data out the socket door, when using UDP, it must first attach a destination address
to the packet. After the packet passes through the sender’s socket, the Internet will
use this destination address to route the packet through the Internet to the socket in
the receiving process. When the packet arrives at the receiving socket, the receiving
process will retrieve the packet through the socket, and then inspect the packet’s
contents and take appropriate action.
So you may be now wondering, what goes into the destination address that is
attached to the packet? As you might expect, the destination host’s IP address is part of
the destination address. By including the destination IP address in the packet, the
routers in the Internet will be able to route the packet through the Internet to the destination host. But because a host may be running many network application processes,
each with one or more sockets, it is also necessary to identify the particular socket in
the destination host. When a socket is created, an identifier, called a port number, is
assigned to it. So, as you might expect, the packet’s destination address also includes
the socket’s port number. In summary, the sending process attaches to the packet a destination address which consists of the destination host’s IP address and the destination
socket’s port number. Moreover, as we shall soon see, the sender’s source address—
consisting of the IP address of the source host and the port number of the source
socket—are also attached to the packet. However, attaching the source address to the
packet is typically not done by the UDP application code; instead it is automatically
done by the underlying operating system.
We’ll use the following simple client-server application to demonstrate socket
programming for both UDP and TCP:
1. The client reads a line of characters (data) from its keyboard and sends the data
to the server.
2. The server receives the data and converts the characters to uppercase.
3. The server sends the modified data to the client.
4. The client receives the modified data and displays the line on its screen.
Figure 2.28 highlights the main socket-related activity of the client and server that
communicate over the UDP transport service.
Now let’s get our hands dirty and take a look at the client-server program pair
for a UDP implementation of this simple application. We also provide a detailed,
line-by-line analysis after each program. We’ll begin with the UDP client, which
will send a simple application-level message to the server. In order for the server to
be able to receive and reply to the client’s message, it must be ready and running—
that is, it must be running as a process before the client sends its message.
The client program is called UDPClient.py, and the server program is called
UDPServer.py. In order to emphasize the key issues, we intentionally provide code
that is minimal. “Good code” would certainly have a few more auxiliary lines, in
2.7
•
SOCKET PROGRAMMING: CREATING NETWORK APPLICATIONS
Server
(Running on serverIP)
Client
Create socket, port=x:
serverSocket =
socket(AF_INET,SOCK_DGRAM)
Create socket:
clientSocket =
socket(AF_INET,SOCK_DGRAM)
Read UDP segment from
serverSocket
Create datagram with serverIP
and port=x ;
send datagram via
clientSocket
Write reply to
serverSocket
specifying client address,
port number
Read datagram from
clientSocket
Close
clientSocket
Figure 2.28
The client-server application using UDP
particular for handling error cases. For this application, we have arbitrarily chosen
12000 for the server port number.
UDPClient.py
Here is the code for the client side of the application:
from socket import *
serverName = ‘hostname’
serverPort = 12000
clientSocket = socket(socket.AF_INET, socket.SOCK_DGRAM)
message = raw_input(’Input lowercase sentence:’)
clientSocket.sendto(message,(serverName, serverPort))
modifiedMessage, serverAddress = clientSocket.recvfrom(2048)
print modifiedMessage
clientSocket.close()
159
160
CHAPTER 2
•
APPLICATION LAYER
Now let’s take a look at the various lines of code in UDPClient.py.
from socket import *
The socket module forms the basis of all network communications in Python. By
including this line, we will be able to create sockets within our program.
serverName = ‘hostname’
serverPort = 12000
The first line sets the string serverName to hostname. Here, we provide a string
containing either the IP address of the server (e.g., “128.138.32.126”) or the hostname of the server (e.g., “cis.poly.edu”). If we use the hostname, then a DNS lookup
will automatically be performed to get the IP address.) The second line sets the integer variable serverPort to 12000.
clientSocket = socket(socket.AF_INET, socket.SOCK_DGRAM)
This line creates the client’s socket, called clientSocket. The first parameter
indicates the address family; in particular, AF_INET indicates that the underlying
network is using IPv4. (Do not worry about this now—we will discuss IPv4 in
Chapter 4.) The second parameter indicates that the socket is of type SOCK_DGRAM,
which means it is a UDP socket (rather than a TCP socket). Note that we are not
specifying the port number of the client socket when we create it; we are instead letting the operating system do this for us. Now that the client process’s door has been
created, we will want to create a message to send through the door.
message = raw_input(’Input lowercase sentence:’)
raw_input() is a built-in function in Python. When this command is executed,
the user at the client is prompted with the words “Input data:” The user then uses her
keyboard to input a line, which is put into the variable message. Now that we have
a socket and a message, we will want to send the message through the socket to the
destination host.
clientSocket.sendto(message,(serverName, serverPort))
In the above line, the method sendto() attaches the destination address
(serverName, serverPort) to the message and sends the resulting packet into
the process’s socket, clientSocket. (As mentioned earlier, the source address is
also attached to the packet, although this is done automatically rather than explicitly
by the code.) Sending a client-to-server message via a UDP socket is that simple!
After sending the packet, the client waits to receive data from the server.
2.7
•
SOCKET PROGRAMMING: CREATING NETWORK APPLICATIONS
modifiedMessage, serverAddress = clientSocket.recvfrom(2048)
With the above line, when a packet arrives from the Internet at the client’s socket,
the packet’s data is put into the variable modifiedMessage and the packet’s
source address is put into the variable serverAddress. The variable
serverAddress contains both the server’s IP address and the server’s port
number. The program UDPClient doesn’t actually need this server address information, since it already knows the server address from the outset; but this line of
Python provides the server address nevertheless. The method recvfrom also
takes the buffer size 2048 as input. (This buffer size works for most purposes.)
print modifiedMessage
This line prints out modifiedMessage on the user’s display. It should be the original
line that the user typed, but now capitalized.
clientSocket.close()
This line closes the socket. The process then terminates.
UDPServer.py
Let’s now take a look at the server side of the application:
from socket import *
serverPort = 12000
serverSocket = socket(AF_INET, SOCK_DGRAM)
serverSocket.bind((’’, serverPort))
print ”The server is ready to receive”
while 1:
message, clientAddress = serverSocket.recvfrom(2048)
modifiedMessage = message.upper()
serverSocket.sendto(modifiedMessage, clientAddress)
Note that the beginning of UDPServer is similar to UDPClient. It also imports the
socket module, also sets the integer variable serverPort to 12000, and also creates a socket of type SOCK_DGRAM (a UDP socket). The first line of code that is
significantly different from UDPClient is:
serverSocket.bind((’’, serverPort))
The above line binds (that is, assigns) the port number 12000 to the server’s socket.
Thus in UDPServer, the code (written by the application developer) is explicitly
161
162
CHAPTER 2
•
APPLICATION LAYER
assigning a port number to the socket. In this manner, when anyone sends a packet
to port 12000 at the IP address of the server, that packet will be directed to this
socket. UDPServer then enters a while loop; the while loop will allow UDPServer
to receive and process packets from clients indefinitely. In the while loop,
UDPServer waits for a packet to arrive.
message, clientAddress = serverSocket.recvfrom(2048)
This line of code is similar to what we saw in UDPClient. When a packet arrives at
the server’s socket, the packet’s data is put into the variable message and the
packet’s source address is put into the variable clientAddress. The variable
clientAddress contains both the client’s IP address and the client’s port number.
Here, UDPServer will make use of this address information, as it provides a return
address, similar to the return address with ordinary postal mail. With this source
address information, the server now knows to where it should direct its reply.
modifiedMessage = message.upper()
This line is the heart of our simple application. It takes the line sent by the client and
uses the method upper() to capitalize it.
serverSocket.sendto(modifiedMessage, clientAddress)
This last line attaches the client’s address (IP address and port number) to the capitalized message, and sends the resulting packet into the server’s socket. (As mentioned earlier, the server address is also attached to the packet, although this is done
automatically rather than explicitly by the code.) The Internet will then deliver the
packet to this client address. After the server sends the packet, it remains in the
while loop, waiting for another UDP packet to arrive (from any client running on
any host).
To test the pair of programs, you install and compile UDPClient.py in one host
and UDPServer.py in another host. Be sure to include the proper hostname or IP
address of the server in UDPClient.py. Next, you execute UDPServer.py, the compiled server program, in the server host. This creates a process in the server that
idles until it is contacted by some client. Then you execute UDPClient.py, the compiled client program, in the client. This creates a process in the client. Finally, to use
the application at the client, you type a sentence followed by a carriage return.
To develop your own UDP client-server application, you can begin by
slightly modifying the client or server programs. For example, instead of converting all the letters to uppercase, the server could count the number of times the
letter s appears and return this number. Or you can modify the client so that after
receiving a capitalized sentence, the user can continue to send more sentences to
the server.
2.7
•
SOCKET PROGRAMMING: CREATING NETWORK APPLICATIONS
2.7.2 Socket Programming with TCP
Unlike UDP, TCP is a connection-oriented protocol. This means that before the client
and server can start to send data to each other, they first need to handshake and establish a TCP connection. One end of the TCP connection is attached to the client socket
and the other end is attached to a server socket. When creating the TCP connection,
we associate with it the client socket address (IP address and port number) and the
server socket address (IP address and port number). With the TCP connection established, when one side wants to send data to the other side, it just drops the data into
the TCP connection via its socket. This is different from UDP, for which the server
must attach a destination address to the packet before dropping it into the socket.
Now let’s take a closer look at the interaction of client and server programs in
TCP. The client has the job of initiating contact with the server. In order for the
server to be able to react to the client’s initial contact, the server has to be ready.
This implies two things. First, as in the case of UDP, the TCP server must be running as a process before the client attempts to initiate contact. Second, the server
program must have a special door—more precisely, a special socket—that welcomes some initial contact from a client process running on an arbitrary host. Using
our house/door analogy for a process/socket, we will sometimes refer to the client’s
initial contact as “knocking on the welcoming door.”
With the server process running, the client process can initiate a TCP connection to the server. This is done in the client program by creating a TCP socket. When
the client creates its TCP socket, it specifies the address of the welcoming socket in
the server, namely, the IP address of the server host and the port number of the
socket. After creating its socket, the client initiates a three-way handshake and
establishes a TCP connection with the server. The three-way handshake, which takes
place within the transport layer, is completely invisible to the client and server programs.
During the three-way handshake, the client process knocks on the welcoming door
of the server process. When the server “hears” the knocking, it creates a new door—
more precisely, a new socket that is dedicated to that particular client. In our example
below, the welcoming door is a TCP socket object that we call serverSocket; the
newly created socket dedicated to the client making the connection is called connectionSocket. Students who are encountering TCP sockets for the first time sometimes confuse the welcoming socket (which is the initial point of contact for all clients
wanting to communicate with the server), and each newly created server-side connection socket that is subsequently created for communicating with each client.
From the application’s perspective, the client’s socket and the server’s connection socket are directly connected by a pipe. As shown in Figure 2.29, the client
process can send arbitrary bytes into its socket, and TCP guarantees that the server
process will receive (through the connection socket) each byte in the order sent. TCP
thus provides a reliable service between the client and server processes. Furthermore,
just as people can go in and out the same door, the client process not only sends bytes
163
164
CHAPTER 2
•
APPLICATION LAYER
Client process
Server process
Th
Client
socket
Figure 2.29
re
e-
y
wa
ha
nd
a
sh
ke
bytes
bytes
Welcoming
socket
Connection
socket
The TCPServer process has two sockets
into but also receives bytes from its socket; similarly, the server process not only
receives bytes from but also sends bytes into its connection socket.
We use the same simple client-server application to demonstrate socket programming with TCP: The client sends one line of data to the server, the server capitalizes
the line and sends it back to the client. Figure 2.30 highlights the main socket-related
activity of the client and server that communicate over the TCP transport service.
TCPClient.py
Here is the code for the client side of the application:
from socket import *
serverName = ’servername’
serverPort = 12000
clientSocket = socket(AF_INET, SOCK_STREAM)
clientSocket.connect((serverName,serverPort))
sentence = raw_input(‘Input lowercase sentence:’)
clientSocket.send(sentence)
modifiedSentence = clientSocket.recv(1024)
print ‘From Server:’, modifiedSentence
clientSocket.close()
2.7
•
SOCKET PROGRAMMING: CREATING NETWORK APPLICATIONS
Client
Server
(Running on serverIP)
Create socket, port=x,
for incoming request:
serverSocket =
socket()
Wait for incoming
connection request:
connectionSocket =
serverSocket.accept()
Read request from
connectionSocket
TCP
connection setup
Create socket, connect
to serverIP, port=x:
clientSocket =
socket()
Send request using
clientSocket
Write reply to
connectionSocket
Close
connectionSocket
Figure 2.30
Read reply from
clientSocket
Close
clientSocket
The client-server application using TCP
Let’s now take a look at the various lines in the code that differ significantly
from the UDP implementation. The first such line is the creation of the client
socket.
clientSocket = socket(AF_INET, SOCK_STREAM)
This line creates the client’s socket, called clientSocket. The first parameter
again indicates that the underlying network is using IPv4. The second parameter
indicates that the socket is of type SOCK_STREAM, which means it is a TCP socket
(rather than a UDP socket). Note that we are again not specifying the port number
165
166
CHAPTER 2
•
APPLICATION LAYER
of the client socket when we create it; we are instead letting the operating system
do this for us. Now the next line of code is very different from what we saw in
UDPClient:
clientSocket.connect((serverName,serverPort))
Recall that before the client can send data to the server (or vice versa) using a TCP
socket, a TCP connection must first be established between the client and server.
The above line initiates the TCP connection between the client and server. The
parameter of the connect() method is the address of the server side of the connection. After this line of code is executed, the three-way handshake is performed
and a TCP connection is established between the client and server.
sentence = raw_input(‘Input lowercase sentence:’)
As with UDPClient, the above obtains a sentence from the user. The string sentence
continues to gather characters until the user ends the line by typing a carriage return.
The next line of code is also very different from UDPClient:
clientSocket.send(sentence)
The above line sends the string sentence through the client’s socket and into the
TCP connection. Note that the program does not explicitly create a packet and attach
the destination address to the packet, as was the case with UDP sockets. Instead the
client program simply drops the bytes in the string sentence into the TCP connection. The client then waits to receive bytes from the server.
modifiedSentence = clientSocket.recv(2048)
When characters arrive from the server, they get placed into the string modifiedSentence. Characters continue to accumulate in modifiedSentence until the
line ends with a carriage return character. After printing the capitalized sentence, we
close the client’s socket:
clientSocket.close()
This last line closes the socket and, hence, closes the TCP connection between the
client and the server. It causes TCP in the client to send a TCP message to TCP in
the server (see Section 3.5).
TCPServer.py
Now let’s take a look at the server program.
2.7
•
SOCKET PROGRAMMING: CREATING NETWORK APPLICATIONS
from socket import *
serverPort = 12000
serverSocket = socket(AF_INET,SOCK_STREAM)
serverSocket.bind((‘’,serverPort))
serverSocket.listen(1)
print ‘The server is ready to receive’
while 1:
connectionSocket, addr = serverSocket.accept()
sentence = connectionSocket.recv(1024)
capitalizedSentence = sentence.upper()
connectionSocket.send(capitalizedSentence)
connectionSocket.close()
Let’s now take a look at the lines that differ significantly from UDPServer and TCPClient. As with TCPClient, the server creates a TCP socket with:
serverSocket=socket(AF_INET,SOCK_STREAM)
Similar to UDPServer, we associate the server port number, serverPort, with
this socket:
serverSocket.bind((‘’,serverPort))
But with TCP, serverSocket will be our welcoming socket. After establishing this welcoming door, we will wait and listen for some client to knock on the
door:
serverSocket.listen(1)
This line has the server listen for TCP connection requests from the client. The
parameter specifies the maximum number of queued connections (at least 1).
connectionSocket, addr = serverSocket.accept()
When a client knocks on this door, the program invokes the accept() method for
serverSocket, which creates a new socket in the server, called connectionSocket, dedicated to this particular client. The client and server then complete
the handshaking, creating a TCP connection between the client’s clientSocket
and the server’s connectionSocket. With the TCP connection established, the
client and server can now send bytes to each other over the connection. With TCP, all
bytes sent from one side not are not only guaranteed to arrive at the other side but also
guaranteed arrive in order.
connectionSocket.close()
167
168
CHAPTER 2
•
APPLICATION LAYER
In this program, after sending the modified sentence to the client, we close the connection socket. But since serverSocket remains open, another client can now
knock on the door and send the server a sentence to modify.
This completes our discussion of socket programming in TCP. You are encouraged to run the two programs in two separate hosts, and also to modify them to
achieve slightly different goals. You should compare the UDP program pair with the
TCP program pair and see how they differ. You should also do many of the socket
programming assignments described at the ends of Chapters 2, 4, and 7. Finally, we
hope someday, after mastering these and more advanced socket programs, you will
write your own popular network application, become very rich and famous, and
remember the authors of this textbook!
2.8 Summary
In this chapter, we’ve studied the conceptual and the implementation aspects of network applications. We’ve learned about the ubiquitous client-server architecture
adopted by many Internet applications and seen its use in the HTTP, FTP, SMTP,
POP3, and DNS protocols. We’ve studied these important application-level protocols, and their corresponding associated applications (the Web, file transfer, e-mail,
and DNS) in some detail. We’ve also learned about the increasingly prevalent P2P
architecture and how it is used in many applications. We’ve examined how the
socket API can be used to build network applications. We’ve walked through the use
of sockets for connection-oriented (TCP) and connectionless (UDP) end-to-end
transport services. The first step in our journey down the layered network architecture is now complete!
At the very beginning of this book, in Section 1.1, we gave a rather vague, barebones definition of a protocol: “the format and the order of messages exchanged
between two or more communicating entities, as well as the actions taken on the transmission and/or receipt of a message or other event.” The material in this chapter, and
in particular our detailed study of the HTTP, FTP, SMTP, POP3, and DNS protocols,
has now added considerable substance to this definition. Protocols are a key concept
in networking; our study of application protocols has now given us the opportunity to
develop a more intuitive feel for what protocols are all about.
In Section 2.1, we described the service models that TCP and UDP offer to
applications that invoke them. We took an even closer look at these service models
when we developed simple applications that run over TCP and UDP in Section 2.7.
However, we have said little about how TCP and UDP provide these service models. For example, we know that TCP provides a reliable data service, but we haven’t
said yet how it does so. In the next chapter we’ll take a careful look at not only the
what, but also the how and why of transport protocols.
HOMEWORK PROBLEMS AND QUESTIONS
Equipped with knowledge about Internet application structure and applicationlevel protocols, we’re now ready to head further down the protocol stack and examine the transport layer in Chapter 3.
Homework Problems and Questions
Chapter 2 Review Questions
SECTION 2.1
R1. List five nonproprietary Internet applications and the application-layer
protocols that they use.
R2. What is the difference between network architecture and application
architecture?
R3. For a communication session between a pair of processes, which process is
the client and which is the server?
R4. For a P2P file-sharing application, do you agree with the statement, “There is
no notion of client and server sides of a communication session”? Why or
why not?
R5. What information is used by a process running on one host to identify a
process running on another host?
R6. Suppose you wanted to do a transaction from a remote client to a server as
fast as possible. Would you use UDP or TCP? Why?
R7. Referring to Figure 2.4, we see that none of the applications listed in Figure
2.4 requires both no data loss and timing. Can you conceive of an application
that requires no data loss and that is also highly time-sensitive?
R8. List the four broad classes of services that a transport protocol can provide.
For each of the service classes, indicate if either UDP or TCP (or both) provides such a service.
R9. Recall that TCP can be enhanced with SSL to provide process-to-process
security services, including encryption. Does SSL operate at the transport
layer or the application layer? If the application developer wants TCP to be
enhanced with SSL, what does the developer have to do?
SECTIONS 2.2–2.5
R10. What is meant by a handshaking protocol?
R11. Why do HTTP, FTP, SMTP, and POP3 run on top of TCP rather than on UDP?
R12. Consider an e-commerce site that wants to keep a purchase record for each of
its customers. Describe how this can be done with cookies.
169
170
CHAPTER 2
•
APPLICATION LAYER
R13. Describe how Web caching can reduce the delay in receiving a requested
object. Will Web caching reduce the delay for all objects requested by a user
or for only some of the objects? Why?
R14. Telnet into a Web server and send a multiline request message. Include in the
request message the If-modified-since: header line to force a
response message with the 304 Not Modified status code.
R15. Why is it said that FTP sends control information “out-of-band”?
R16. Suppose Alice, with a Web-based e-mail account (such as Hotmail or gmail),
sends a message to Bob, who accesses his mail from his mail server using
POP3. Discuss how the message gets from Alice’s host to Bob’s host. Be sure
to list the series of application-layer protocols that are used to move the message between the two hosts.
R17. Print out the header of an e-mail message you have recently received. How
many Received: header lines are there? Analyze each of the header lines
in the message.
R18. From a user’s perspective, what is the difference between the download-anddelete mode and the download-and-keep mode in POP3?
R19. Is it possible for an organization’s Web server and mail server to have exactly
the same alias for a hostname (for example, foo.com)? What would be the
type for the RR that contains the hostname of the mail server?
R20. Look over your received emails, and examine the header of a message sent
from a user with an .edu email address. Is it possible to determine from the
header the IP address of the host from which the message was sent? Do the
same for a message sent from a gmail account.
SECTION 2.6
R21. In BitTorrent, suppose Alice provides chunks to Bob throughout a 30-second
interval. Will Bob necessarily return the favor and provide chunks to Alice in
this same interval? Why or why not?
R22. Consider a new peer Alice that joins BitTorrent without possessing any
chunks. Without any chunks, she cannot become a top-four uploader for any
of the other peers, since she has nothing to upload. How then will Alice get
her first chunk?
R23. What is an overlay network? Does it include routers? What are the edges in
the overlay network?
R24. Consider a DHT with a mesh overlay topology (that is, every peer tracks all
peers in the system). What are the advantages and disadvantages of such a
design? What are the advantages and disadvantages of a circular DHT (with
no shortcuts)?
PROBLEMS
R25. List at least four different applications that are naturally suitable for P2P
architectures. (Hint: File distribution and instant messaging are two.)
SECTION 2.7
R26. In Section 2.7, the UDP server described needed only one socket, whereas the
TCP server needed two sockets. Why? If the TCP server were to support n
simultaneous connections, each from a different client host, how many sockets
would the TCP server need?
R27. For the client-server application over TCP described in Section 2.7, why must
the server program be executed before the client program? For the clientserver application over UDP, why may the client program be executed before
the server program?
Problems
P1. True or false?
a. A user requests a Web page that consists of some text and three images.
For this page, the client will send one request message and receive four
response messages.
b. Two distinct Web pages (for example, www.mit.edu/research.html
and www.mit.edu/students.html) can be sent over the same persistent connection.
c. With nonpersistent connections between browser and origin server, it is possible for a single TCP segment to carry two distinct HTTP request messages.
d. The Date: header in the HTTP response message indicates when the
object in the response was last modified.
e. HTTP response messages never have an empty message body.
P2. Read RFC 959 for FTP. List all of the client commands that are supported by
the RFC.
P3. Consider an HTTP client that wants to retrieve a Web document at a given
URL. The IP address of the HTTP server is initially unknown. What transport
and application-layer protocols besides HTTP are needed in this scenario?
P4. Consider the following string of ASCII characters that were captured by
Wireshark when the browser sent an HTTP GET message (i.e., this is the actual
content of an HTTP GET message). The characters <cr><lf> are carriage
return and line-feed characters (that is, the italized character string <cr> in
the text below represents the single carriage-return character that was contained at that point in the HTTP header). Answer the following questions,
indicating where in the HTTP GET message below you find the answer.
171
172
CHAPTER 2
•
APPLICATION LAYER
GET /cs453/index.html HTTP/1.1<cr><lf>Host: gai
a.cs.umass.edu<cr><lf>User-Agent: Mozilla/5.0 (
Windows;U; Windows NT 5.1; en-US; rv:1.7.2) Gec
ko/20040804 Netscape/7.2 (ax) <cr><lf>Accept:ex
t/xml, application/xml, application/xhtml+xml, text
/html;q=0.9, text/plain;q=0.8,image/png,*/*;q=0.5
<cr><lf>Accept-Language: en-us,en;q=0.5<cr><lf>AcceptEncoding: zip,deflate<cr><lf>Accept-Charset: ISO
-8859-1,utf-8;q=0.7,*;q=0.7<cr><lf>Keep-Alive: 300<cr>
<lf>Connection:keep-alive<cr><lf><cr><lf>
a.
b.
c.
d.
e.
What is the URL of the document requested by the browser?
What version of HTTP is the browser running?
Does the browser request a non-persistent or a persistent connection?
What is the IP address of the host on which the browser is running?
What type of browser initiates this message? Why is the browser type
needed in an HTTP request message?
P5. The text below shows the reply sent from the server in response to the HTTP
GET message in the question above. Answer the following questions, indicating where in the message below you find the answer.
HTTP/1.1 200 OK<cr><lf>Date: Tue, 07 Mar 2008
12:39:45GMT<cr><lf>Server: Apache/2.0.52 (Fedora)
<cr><lf>Last-Modified: Sat, 10 Dec2005 18:27:46
GMT<cr><lf>ETag: “526c3-f22-a88a4c80”<cr><lf>AcceptRanges: bytes<cr><lf>Content-Length: 3874<cr><lf>
Keep-Alive: timeout=max=100<cr><lf>Connection:
Keep-Alive<cr><lf>Content-Type: text/html; charset=
ISO-8859-1<cr><lf><cr><lf><!doctype html public “//w3c//dtd html 4.0 transitional//en”><lf><html><lf>
<head><lf> <meta http-equiv=”Content-Type”
content=”text/html; charset=iso-8859-1”><lf> <meta
name=”GENERATOR” content=”Mozilla/4.79 [en] (Windows NT
5.0; U) Netscape]”><lf> <title>CMPSCI 453 / 591 /
NTU-ST550A Spring 2005 homepage</title><lf></head><lf>
<much more document text following here (not shown)>
a. Was the server able to successfully find the document or not? What time
was the document reply provided?
b. When was the document last modified?
c. How many bytes are there in the document being returned?
d. What are the first 5 bytes of the document being returned? Did the server
agree to a persistent connection?
PROBLEMS
P6. Obtain the HTTP/1.1 specification (RFC 2616). Answer the following questions:
a. Explain the mechanism used for signaling between the client and server
to indicate that a persistent connection is being closed. Can the client, the
server, or both signal the close of a connection?
b. What encryption services are provided by HTTP?
c. Can a client open three or more simultaneous connections with a given
server?
d. Either a server or a client may close a transport connection between them
if either one detects the connection has been idle for some time. Is it possible that one side starts closing a connection while the other side is transmitting data via this connection? Explain.
P7. Suppose within your Web browser you click on a link to obtain a Web page.
The IP address for the associated URL is not cached in your local host, so a
DNS lookup is necessary to obtain the IP address. Suppose that n DNS
servers are visited before your host receives the IP address from DNS; the
successive visits incur an RTT of RTT1, . . ., RTTn. Further suppose that the
Web page associated with the link contains exactly one object, consisting of a
small amount of HTML text. Let RTT0 denote the RTT between the local host
and the server containing the object. Assuming zero transmission time of the
object, how much time elapses from when the client clicks on the link until
the client receives the object?
P8. Referring to Problem P7, suppose the HTML file references eight very small
objects on the same server. Neglecting transmission times, how much time
elapses with
a. Non-persistent HTTP with no parallel TCP connections?
b. Non-persistent HTTP with the browser configured for 5 parallel connections?
c. Persistent HTTP?
P9. Consider Figure 2.12, for which there is an institutional network connected to
the Internet. Suppose that the average object size is 850,000 bits and that the
average request rate from the institution’s browsers to the origin servers is 16
requests per second. Also suppose that the amount of time it takes from when
the router on the Internet side of the access link forwards an HTTP request
until it receives the response is three seconds on average (see Section 2.2.5).
Model the total average response time as the sum of the average access delay
(that is, the delay from Internet router to institution router) and the average
Internet delay. For the average access delay, use Δ/(1 – Δ␤), where Δ is the
average time required to send an object over the access link and ␤ is the
arrival rate of objects to the access link.
a. Find the total average response time.
b. Now suppose a cache is installed in the institutional LAN. Suppose the
miss rate is 0.4. Find the total response time.
173
174
CHAPTER 2
•
APPLICATION LAYER
P10. Consider a short, 10-meter link, over which a sender can transmit at a rate of
150 bits/sec in both directions. Suppose that packets containing data are
100,000 bits long, and packets containing only control (e.g., ACK or handshaking) are 200 bits long. Assume that N parallel connections each get 1/N
of the link bandwidth. Now consider the HTTP protocol, and suppose that
each downloaded object is 100 Kbits long, and that the initial downloaded
object contains 10 referenced objects from the same sender. Would parallel
downloads via parallel instances of non-persistent HTTP make sense in this
case? Now consider persistent HTTP. Do you expect significant gains over
the non-persistent case? Justify and explain your answer.
P11. Consider the scenario introduced in the previous problem. Now suppose that
the link is shared by Bob with four other users. Bob uses parallel instances of
non-persistent HTTP, and the other four users use non-persistent HTTP without parallel downloads.
a. Do Bob’s parallel connections help him get Web pages more quickly?
Why or why not?
b. If all five users open five parallel instances of non-persistent HTTP, then
would Bob’s parallel connections still be beneficial? Why or why not?
P12. Write a simple TCP program for a server that accepts lines of input from a
client and prints the lines onto the server’s standard output. (You can do this by
modifying the TCPServer.py program in the text.) Compile and execute your
program. On any other machine that contains a Web browser, set the proxy
server in the browser to the host that is running your server program; also configure the port number appropriately. Your browser should now send its GET
request messages to your server, and your server should display the messages
on its standard output. Use this platform to determine whether your browser
generates conditional GET messages for objects that are locally cached.
P13. What is the difference between MAIL FROM: in SMTP and From: in the
mail message itself?
P14. How does SMTP mark the end of a message body? How about HTTP? Can
HTTP use the same method as SMTP to mark the end of a message body?
Explain.
P15. Read RFC 5321 for SMTP. What does MTA stand for? Consider the following received spam email (modified from a real spam email). Assuming only
the originator of this spam email is malacious and all other hosts are honest,
identify the malacious host that has generated this spam email.
From - Fri Nov 07 13:41:30 2008
Return-Path: <[email protected]>
Received: from barmail.cs.umass.edu
(barmail.cs.umass.edu [128.119.240.3]) by cs.umass.edu
(8.13.1/8.12.6) for <[email protected]>; Fri, 7 Nov 2008
13:27:10 -0500
PROBLEMS
Received: from asusus-4b96 (localhost [127.0.0.1]) by
barmail.cs.umass.edu (Spam Firewall) for
<[email protected]>; Fri, 7 Nov 2008 13:27:07 -0500
(EST)
Received: from asusus-4b96 ([58.88.21.177]) by
barmail.cs.umass.edu for <[email protected]>; Fri,
07 Nov 2008 13:27:07 -0500 (EST)
Received: from [58.88.21.177] by
inbnd55.exchangeddd.com; Sat, 8 Nov 2008 01:27:07 +0700
From: "Jonny" <[email protected]>
To: <[email protected]>
Subject: How to secure your savings
P16. Read the POP3 RFC, RFC 1939. What is the purpose of the UIDL POP3
command?
P17. Consider accessing your e-mail with POP3.
a. Suppose you have configured your POP mail client to operate in the
download-and-delete mode. Complete the following transaction:
C:
S:
S:
S:
C:
S:
S:
S:
?
?
list
1 498
2 912
.
retr 1
blah blah ...
..........blah
.
b. Suppose you have configured your POP mail client to operate in the
download-and-keep mode. Complete the following transaction:
C:
S:
S:
S:
C:
S:
S:
S:
?
?
list
1 498
2 912
.
retr 1
blah blah ...
..........blah
.
175
176
CHAPTER 2
•
APPLICATION LAYER
c. Suppose you have configured your POP mail client to operate in the
download-and-keep mode. Using your transcript in part (b), suppose you
retrieve messages 1 and 2, exit POP, and then five minutes later you again
access POP to retrieve new e-mail. Suppose that in the five-minute interval no new messages have been sent to you. Provide a transcript of this
second POP session.
P18. a. What is a whois database?
b. Use various whois databases on the Internet to obtain the names of two
DNS servers. Indicate which whois databases you used.
c. Use nslookup on your local host to send DNS queries to three DNS servers:
your local DNS server and the two DNS servers you found in part (b). Try
querying for Type A, NS, and MX reports. Summarize your findings.
d. Use nslookup to find a Web server that has multiple IP addresses. Does
the Web server of your institution (school or company) have multiple IP
addresses?
e. Use the ARIN whois database to determine the IP address range used by
your university.
f. Describe how an attacker can use whois databases and the nslookup tool
to perform reconnaissance on an institution before launching an attack.
g. Discuss why whois databases should be publicly available.
P19. In this problem, we use the useful dig tool available on Unix and Linux hosts
to explore the hierarchy of DNS servers. Recall that in Figure 2.21, a DNS
server higher in the DNS hierarchy delegates a DNS query to a DNS server
lower in the hierarchy, by sending back to the DNS client the name of that
lower-level DNS server. First read the man page for dig, and then answer the
following questions.
a. Starting with a root DNS server (from one of the root servers [a-m].rootservers.net), initiate a sequence of queries for the IP address for your
department’s Web server by using dig. Show the list of the names of DNS
servers in the delegation chain in answering your query.
b. Repeat part a) for several popular Web sites, such as google.com,
yahoo.com, or amazon.com.
P20. Suppose you can access the caches in the local DNS servers of your department.
Can you propose a way to roughly determine the Web servers (outside your
department) that are most popular among the users in your department? Explain.
P21. Suppose that your department has a local DNS server for all computers in the
department. You are an ordinary user (i.e., not a network/system administrator). Can you determine if an external Web site was likely accessed from a
computer in your department a couple of seconds ago? Explain.
PROBLEMS
P22. Consider distributing a file of F = 15 Gbits to N peers. The server has an upload
rate of us = 30 Mbps, and each peer has a download rate of di = 2 Mbps and an
upload rate of u. For N = 10, 100, and 1,000 and u = 300 Kbps, 700 Kbps, and
2 Mbps, prepare a chart giving the minimum distribution time for each of
the combinations of N and u for both client-server distribution and P2P
distribution.
P23. Consider distributing a file of F bits to N peers using a client-server architecture. Assume a fluid model where the server can simultaneously transmit to
multiple peers, transmitting to each peer at different rates, as long as the combined rate does not exceed us.
a. Suppose that us/N ≤ dmin. Specify a distribution scheme that has a distribution time of NF/us.
b. Suppose that us/N ≥ dmin. Specify a distribution scheme that has a distribution time of F/ dmin.
c. Conclude that the minimum distribution time is in general given by
max{NF/us, F/ dmin}.
P24. Consider distributing a file of F bits to N peers using a P2P architecture.
Assume a fluid model. For simplicity assume that dmin is very large, so that
peer download bandwidth is never a bottleneck.
a. Suppose that us ≤ (us + u1 + ... + uN)/N. Specify a distribution scheme
that has a distribution time of F/us.
b. Suppose that us ≥ (us + u1 + ... + uN)/N. Specify a distribution scheme
that has a distribution time of NF/(us + u1 + ... + uN).
c. Conclude that the minimum distribution time is in general given by
max{F/us, NF/(us + u1 + ... + uN)}.
P25. Consider an overlay network with N active peers, with each pair of peers having an active TCP connection. Additionally, suppose that the TCP connections
pass through a total of M routers. How many nodes and edges are there in the
corresponding overlay network?
P26. Suppose Bob joins a BitTorrent torrent, but he does not want to upload any
data to any other peers (so called free-riding).
a. Bob claims that he can receive a complete copy of the file that is shared
by the swarm. Is Bob’s claim possible? Why or why not?
b. Bob further claims that he can further make his “free-riding” more efficient by using a collection of multiple computers (with distinct IP
addresses) in the computer lab in his department. How can he do that?
P27. In the circular DHT example in Section 2.6.2, suppose that peer 3 learns that
peer 5 has left. How does peer 3 update its successor state information?
Which peer is now its first successor? Its second successor?
177
VideoNote
Walking through
distributed hash tables
178
CHAPTER 2
•
APPLICATION LAYER
P28. In the circular DHT example in Section 2.6.2, suppose that a new peer 6
wants to join the DHT and peer 6 initially only knows peer 15’s IP address.
What steps are taken?
P29. Because an integer in [0, 2n ⫺ 1] can be expressed as an n-bit binary number in
a DHT, each key can be expressed as k = (k0, k1, . . . , kn–1), and each peer identifier can be expressed p = (p0, p1, . . . , pn–1). Let’s now define the XOR distance between a key k and peer p as
n-1
d 1 k, p 2 = a 兩 k j - pj 兩2j
j=0
Describe how this metric can be used to assign (key, value) pairs to peers.
(To learn about how to build an efficient DHT using this natural metric, see
[Maymounkov 2002] in which the Kademlia DHT is described.)
P30. As DHTs are overlay networks, they may not necessarily match the underlay physical network well in the sense that two neighboring peers might be
physically very far away; for example, one peer could be in Asia and its
neighbor could be in North America. If we randomly and uniformly assign
identifiers to newly joined peers, would this assignment scheme cause such
a mismatch? Explain. And how would such a mismatch affect the DHT’s
performance?
P31. Install and compile the Python programs TCPClient and UDPClient on one
host and TCPServer and UDPServer on another host.
a. Suppose you run TCPClient before you run TCPServer. What happens?
Why?
b. Suppose you run UDPClient before you run UDPServer. What happens?
Why?
c. What happens if you use different port numbers for the client and server
sides?
P32. Suppose that in UDPClient.py, after we create the socket, we add the line:
clientSocket.bind(('', 5432))
Will it become necessary to change UDPServer.py? What are the port numbers for the sockets in UDPClient and UDPServer? What were they before
making this change?
P33. Can you configure your browser to open multiple simultaneous connections
to a Web site? What are the advantages and disadvantages of having a large
number of simultaneous TCP connections?
P34 We have seen that Internet TCP sockets treat the data being sent as a byte
stream but UDP sockets recognize message boundaries. What are one
SOCKET PROGRAMMING ASSIGNMENTS
advantage and one disadvantage of byte-oriented API versus having the API
explicitly recognize and preserve application-defined message boundaries?
P35. What is the Apache Web server? How much does it cost? What functionality
does it currently have? You may want to look at Wikipedia to answer this
question.
P36. Many BitTorrent clients use DHTs to create a distributed tracker. For these
DHTs, what is the “key” and what is the “value”?
Socket Programming Assignments
The companion Web site includes six socket programming assignments. The first
four assignments are summarized below. The fifth assignment makes use of the
ICMP protocol and is summarized at the end of Chapter 4. The sixth assignment
employs multimedia protocols and is summarized at the end of Chapter 7. It is
highly recommended that students complete several, if not all, of these assignments.
Students can find full details of these assignments, as well as important snippets of
the Python code, at the Web site http://www.awl.com/kurose-ross.
Assignment 1: Web Server
In this assignment, you will develop a simple Web server in Python that is capable
of processing only one request. Specifically, your Web server will (i) create a connection socket when contacted by a client (browser); (ii) receive the HTTP request
from this connection; (iii) parse the request to determine the specific file being
requested; (iv) get the requested file from the server’s file system; (v) create an
HTTP response message consisting of the requested file preceded by header lines;
and (vi) send the response over the TCP connection to the requesting browser. If a
browser requests a file that is not present in your server, your server should return a
“404 Not Found” error message.
In the companion Web site, we provide the skeleton code for your server. Your
job is to complete the code, run your server, and then test your server by sending
requests from browsers running on different hosts. If you run your server on a host
that already has a Web server running on it, then you should use a different port than
port 80 for your Web server.
Assignment 2: UDP Pinger
In this programming assignment, you will write a client ping program in Python.
Your client will send a simple ping message to a server, receive a corresponding
pong message back from the server, and determine the delay between when the
client sent the ping message and received the pong message. This delay is called the
Round Trip Time (RTT). The functionality provided by the client and server is
179
180
CHAPTER 2
•
APPLICATION LAYER
similar to the functionality provided by standard ping program available in modern
operating systems. However, standard ping programs use the Internet Control Message Protocol (ICMP) (which we will study in Chapter 4). Here we will create a
nonstandard (but simple!) UDP-based ping program.
Your ping program is to send 10 ping messages to the target server over UDP.
For each message, your client is to determine and print the RTT when the corresponding pong message is returned. Because UDP is an unreliable protocol, a packet sent
by the client or server may be lost. For this reason, the client cannot wait indefinitely
for a reply to a ping message. You should have the client wait up to one second for a
reply from the server; if no reply is received, the client should assume that the
packet was lost and print a message accordingly.
In this assignment, you will be given the complete code for the server (available in the companion Web site). Your job is to write the client code, which will be
very similar to the server code. It is recommended that you first study carefully the
server code. You can then write your client code, liberally cutting and pasting lines
from the server code.
Assignment 3: Mail Client
The goal of this programming assignment is to create a simple mail client that sends
email to any recipient. Your client will need to establish a TCP connection with a
mail server (e.g., a Google mail server), dialogue with the mail server using the
SMTP protocol, send an email message to a recipient (e.g., your friend) via the mail
server, and finally close the TCP connection with the mail server.
For this assignment, the companion Web site provides the skeleton code for
your client. Your job is to complete the code and test your client by sending
email to different user accounts. You may also try sending through different
servers (for example, through a Google mail server and through your university
mail server).
Assignment 4: Multi-Threaded Web Proxy
In this assignment, you will develop a Web proxy. When your proxy receives an
HTTP request for an object from a browser, it generates a new HTTP request for
the same object and sends it to the origin server. When the proxy receives the
corresponding HTTP response with the object from the origin server, it creates a
new HTTP response, including the object, and sends it to the client. This proxy
will be multi-threaded, so that it will be able to handle multiple requests at the
same time.
For this assignment, the companion Web site provides the skeleton code for the
proxy server. Your job is to complete the code, and then test it by having different
browsers request Web objects via your proxy.
WIRESHARK LABS
181
Wireshark Lab: HTTP
Having gotten our feet wet with the Wireshark packet sniffer in Lab 1, we’re now
ready to use Wireshark to investigate protocols in operation. In this lab, we’ll explore
several aspects of the HTTP protocol: the basic GET/reply interaction, HTTP message
formats, retrieving large HTML files, retrieving HTML files with embedded URLs,
persistent and non-persistent connections, and HTTP authentication and security.
As is the case with all Wireshark labs, the full description of this lab is available
at this book’s Web site, http://www.awl.com/kurose-ross.
Wireshark Lab: DNS
In this lab, we take a closer look at the client side of the DNS, the protocol that translates Internet hostnames to IP addresses. Recall from Section 2.5 that the client’s role in
the DNS is relatively simple—a client sends a query to its local DNS server and
receives a response back. Much can go on under the covers, invisible to the DNS
clients, as the hierarchical DNS servers communicate with each other to either recursively or iteratively resolve the client’s DNS query. From the DNS client’s standpoint,
however, the protocol is quite simple—a query is formulated to the local DNS server
and a response is received from that server. We observe DNS in action in this lab.
As is the case with all Wireshark labs, the full description of this lab is available at
this book’s Web site, http://www.awl.com/kurose-ross.
VideoNote
Using Wireshark to
investigate the
HTTP protocol
AN INTERVIEW WITH...
Marc Andreessen
Marc Andreessen is the co-creator of Mosaic, the Web browser that
popularized the World Wide Web in 1993. Mosaic had a clean,
easily understood interface and was the first browser to display
images in-line with text. In 1994, Marc Andreessen and Jim Clark
founded Netscape, whose browser was by far the most popular
browser through the mid-1990s. Netscape also developed the Secure
Sockets Layer (SSL) protocol and many Internet server products, including mail servers and SSL-based Web servers. He is now a co-founder
and general partner of venture capital firm Andreessen Horowitz, overseeing portfolio development with holdings that include Facebook,
Foursquare, Groupon, Jawbone, Twitter, and Zynga. He serves on
numerous boards, including Bump, eBay, Glam Media, Facebook,
and Hewlett-Packard. He holds a BS in Computer Science from the
University of Illinois at Urbana-Champaign.
How did you become interested in computing? Did you always know that you wanted to
work in information technology?
The video game and personal computing revolutions hit right when I was growing up—
personal computing was the new technology frontier in the late 70’s and early 80’s. And it
wasn’t just Apple and the IBM PC, but hundreds of new companies like Commodore and
Atari as well. I taught myself to program out of a book called “Instant Freeze-Dried BASIC”
at age 10, and got my first computer (a TRS-80 Color Computer—look it up!) at age 12.
Please describe one or two of the most exciting projects you have worked on during your
career. What were the biggest challenges?
Undoubtedly the most exciting project was the original Mosaic web browser in ’92–’93—
and the biggest challenge was getting anyone to take it seriously back then. At the time,
everyone thought the interactive future would be delivered as “interactive television” by
huge companies, not as the Internet by startups.
What excites you about the future of networking and the Internet? What are your biggest
concerns?
The most exciting thing is the huge unexplored frontier of applications and services that
programmers and entrepreneurs are able to explore—the Internet has unleashed creativity at
182
a level that I don’t think we’ve ever seen before. My biggest concern is the principle of
unintended consequences—we don’t always know the implications of what we do, such as
the Internet being used by governments to run a new level of surveillance on citizens.
Is there anything in particular students should be aware of as Web technology advances?
The rate of change—the most important thing to learn is how to learn—how to flexibly
adapt to changes in the specific technologies, and how to keep an open mind on the new
opportunities and possibilities as you move through your career.
What people inspired you professionally?
Vannevar Bush, Ted Nelson, Doug Engelbart, Nolan Bushnell, Bill Hewlett and Dave
Packard, Ken Olsen, Steve Jobs, Steve Wozniak, Andy Grove, Grace Hopper, Hedy Lamarr,
Alan Turing, Richard Stallman.
What are your recommendations for students who want to pursue careers in computing
and information technology?
Go as deep as you possibly can on understanding how technology is created, and then complement with learning how business works.
Can technology solve the world’s problems?
No, but we advance the standard of living of people through economic growth, and most
economic growth throughout history has come from technology—so that’s as good as it gets.
183
This page intentionally left blank
CHAPTER
3
Transport
Layer
Residing between the application and network layers, the transport layer is a central
piece of the layered network architecture. It has the critical role of providing communication services directly to the application processes running on different hosts.
The pedagogic approach we take in this chapter is to alternate between discussions
of transport-layer principles and discussions of how these principles are implemented in existing protocols; as usual, particular emphasis will be given to Internet
protocols, in particular the TCP and UDP transport-layer protocols.
We’ll begin by discussing the relationship between the transport and network
layers. This sets the stage for examining the first critical function of the transport
layer—extending the network layer’s delivery service between two end systems to a
delivery service between two application-layer processes running on the end systems. We’ll illustrate this function in our coverage of the Internet’s connectionless
transport protocol, UDP.
We’ll then return to principles and confront one of the most fundamental problems in computer networking—how two entities can communicate reliably over a
medium that may lose and corrupt data. Through a series of increasingly complicated (and realistic!) scenarios, we’ll build up an array of techniques that transport
protocols use to solve this problem. We’ll then show how these principles are
embodied in TCP, the Internet’s connection-oriented transport protocol.
We’ll next move on to a second fundamentally important problem in networking—
controlling the transmission rate of transport-layer entities in order to avoid, or
185
186
CHAPTER 3
•
TRANSPORT LAYER
recover from, congestion within the network. We’ll consider the causes and consequences of congestion, as well as commonly used congestion-control techniques.
After obtaining a solid understanding of the issues behind congestion control, we’ll
study TCP’s approach to congestion control.
3.1 Introduction and Transport-Layer Services
In the previous two chapters we touched on the role of the transport layer and the
services that it provides. Let’s quickly review what we have already learned about
the transport layer.
A transport-layer protocol provides for logical communication between application processes running on different hosts. By logical communication, we mean
that from an application’s perspective, it is as if the hosts running the processes were
directly connected; in reality, the hosts may be on opposite sides of the planet, connected via numerous routers and a wide range of link types. Application processes
use the logical communication provided by the transport layer to send messages to
each other, free from the worry of the details of the physical infrastructure used to
carry these messages. Figure 3.1 illustrates the notion of logical communication.
As shown in Figure 3.1, transport-layer protocols are implemented in the end
systems but not in network routers. On the sending side, the transport layer converts
the application-layer messages it receives from a sending application process into
transport-layer packets, known as transport-layer segments in Internet terminology.
This is done by (possibly) breaking the application messages into smaller chunks and
adding a transport-layer header to each chunk to create the transport-layer segment.
The transport layer then passes the segment to the network layer at the sending end
system, where the segment is encapsulated within a network-layer packet (a datagram) and sent to the destination. It’s important to note that network routers act only
on the network-layer fields of the datagram; that is, they do not examine the fields of
the transport-layer segment encapsulated with the datagram. On the receiving side,
the network layer extracts the transport-layer segment from the datagram and passes
the segment up to the transport layer. The transport layer then processes the received
segment, making the data in the segment available to the receiving application.
More than one transport-layer protocol may be available to network applications.
For example, the Internet has two protocols—TCP and UDP. Each of these protocols
provides a different set of transport-layer services to the invoking application.
3.1.1 Relationship Between Transport and Network Layers
Recall that the transport layer lies just above the network layer in the protocol stack.
Whereas a transport-layer protocol provides logical communication between
processes running on different hosts, a network-layer protocol provides logical
3.1
•
INTRODUCTION AND TRANSPORT-LAYER SERVICES
187
National or
Global ISP
Mobile Network
Network
Data link
Physical
Network
Data link
Physical
Application
Network
Local or
Regional ISP
Transport
Network
Home Network
Data link
Lo
Physical
gic
Data link
Physical
Network
al
en
d-t
Data link
o-e
nd
Physical
tra
ns
po
rt
Network
Data link
Physical
Application
Enterprise Network
Transport
Network
Data link
Physical
Figure 3.1
The transport layer provides logical rather than physical
communication between application processes
188
CHAPTER 3
•
TRANSPORT LAYER
communication between hosts. This distinction is subtle but important. Let’s examine this distinction with the aid of a household analogy.
Consider two houses, one on the East Coast and the other on the West Coast, with
each house being home to a dozen kids. The kids in the East Coast household are
cousins of the kids in the West Coast household. The kids in the two households love
to write to each other—each kid writes each cousin every week, with each letter delivered by the traditional postal service in a separate envelope. Thus, each household
sends 144 letters to the other household every week. (These kids would save a lot of
money if they had e-mail!) In each of the households there is one kid—Ann in the
West Coast house and Bill in the East Coast house—responsible for mail collection
and mail distribution. Each week Ann visits all her brothers and sisters, collects the
mail, and gives the mail to a postal-service mail carrier, who makes daily visits to the
house. When letters arrive at the West Coast house, Ann also has the job of distributing the mail to her brothers and sisters. Bill has a similar job on the East Coast.
In this example, the postal service provides logical communication between the
two houses—the postal service moves mail from house to house, not from person to
person. On the other hand, Ann and Bill provide logical communication among the
cousins—Ann and Bill pick up mail from, and deliver mail to, their brothers and sisters. Note that from the cousins’ perspective, Ann and Bill are the mail service, even
though Ann and Bill are only a part (the end-system part) of the end-to-end delivery
process. This household example serves as a nice analogy for explaining how the
transport layer relates to the network layer:
application messages = letters in envelopes
processes = cousins
hosts (also called end systems) = houses
transport-layer protocol = Ann and Bill
network-layer protocol = postal service (including mail carriers)
Continuing with this analogy, note that Ann and Bill do all their work within
their respective homes; they are not involved, for example, in sorting mail in any
intermediate mail center or in moving mail from one mail center to another. Similarly, transport-layer protocols live in the end systems. Within an end system, a
transport protocol moves messages from application processes to the network edge
(that is, the network layer) and vice versa, but it doesn’t have any say about how the
messages are moved within the network core. In fact, as illustrated in Figure 3.1,
intermediate routers neither act on, nor recognize, any information that the transport
layer may have added to the application messages.
Continuing with our family saga, suppose now that when Ann and Bill go on
vacation, another cousin pair—say, Susan and Harvey—substitute for them and provide the household-internal collection and delivery of mail. Unfortunately for the
two families, Susan and Harvey do not do the collection and delivery in exactly the
same way as Ann and Bill. Being younger kids, Susan and Harvey pick up and drop
off the mail less frequently and occasionally lose letters (which are sometimes
3.1
•
INTRODUCTION AND TRANSPORT-LAYER SERVICES
chewed up by the family dog). Thus, the cousin-pair Susan and Harvey do not provide the same set of services (that is, the same service model) as Ann and Bill. In an
analogous manner, a computer network may make available multiple transport protocols, with each protocol offering a different service model to applications.
The possible services that Ann and Bill can provide are clearly constrained by
the possible services that the postal service provides. For example, if the postal service doesn’t provide a maximum bound on how long it can take to deliver mail
between the two houses (for example, three days), then there is no way that Ann and
Bill can guarantee a maximum delay for mail delivery between any of the cousin
pairs. In a similar manner, the services that a transport protocol can provide are often
constrained by the service model of the underlying network-layer protocol. If the
network-layer protocol cannot provide delay or bandwidth guarantees for transportlayer segments sent between hosts, then the transport-layer protocol cannot provide
delay or bandwidth guarantees for application messages sent between processes.
Nevertheless, certain services can be offered by a transport protocol even when
the underlying network protocol doesn’t offer the corresponding service at the network layer. For example, as we’ll see in this chapter, a transport protocol can offer
reliable data transfer service to an application even when the underlying network
protocol is unreliable, that is, even when the network protocol loses, garbles, or
duplicates packets. As another example (which we’ll explore in Chapter 8 when we
discuss network security), a transport protocol can use encryption to guarantee that
application messages are not read by intruders, even when the network layer cannot
guarantee the confidentiality of transport-layer segments.
3.1.2 Overview of the Transport Layer in the Internet
Recall that the Internet, and more generally a TCP/IP network, makes two distinct
transport-layer protocols available to the application layer. One of these protocols is
UDP (User Datagram Protocol), which provides an unreliable, connectionless service
to the invoking application. The second of these protocols is TCP (Transmission Control Protocol), which provides a reliable, connection-oriented service to the invoking
application. When designing a network application, the application developer must
specify one of these two transport protocols. As we saw in Section 2.7, the application
developer selects between UDP and TCP when creating sockets.
To simplify terminology, when in an Internet context, we refer to the transportlayer packet as a segment. We mention, however, that the Internet literature (for example, the RFCs) also refers to the transport-layer packet for TCP as a segment but often
refers to the packet for UDP as a datagram. But this same Internet literature also uses
the term datagram for the network-layer packet! For an introductory book on computer
networking such as this, we believe that it is less confusing to refer to both TCP and
UDP packets as segments, and reserve the term datagram for the network-layer packet.
Before proceeding with our brief introduction of UDP and TCP, it will be useful to say a few words about the Internet’s network layer. (We’ll learn about the network layer in detail in Chapter 4.) The Internet’s network-layer protocol has a
189
190
CHAPTER 3
•
TRANSPORT LAYER
name—IP, for Internet Protocol. IP provides logical communication between hosts.
The IP service model is a best-effort delivery service. This means that IP makes its
“best effort” to deliver segments between communicating hosts, but it makes no
guarantees. In particular, it does not guarantee segment delivery, it does not guarantee orderly delivery of segments, and it does not guarantee the integrity of the data
in the segments. For these reasons, IP is said to be an unreliable service. We also
mention here that every host has at least one network-layer address, a so-called IP
address. We’ll examine IP addressing in detail in Chapter 4; for this chapter we need
only keep in mind that each host has an IP address.
Having taken a glimpse at the IP service model, let’s now summarize the service models provided by UDP and TCP. The most fundamental responsibility of UDP
and TCP is to extend IP’s delivery service between two end systems to a delivery
service between two processes running on the end systems. Extending host-to-host
delivery to process-to-process delivery is called transport-layer multiplexing and
demultiplexing. We’ll discuss transport-layer multiplexing and demultiplexing in
the next section. UDP and TCP also provide integrity checking by including errordetection fields in their segments’ headers. These two minimal transport-layer services—process-to-process data delivery and error checking—are the only two
services that UDP provides! In particular, like IP, UDP is an unreliable service—it
does not guarantee that data sent by one process will arrive intact (or at all!) to the
destination process. UDP is discussed in detail in Section 3.3.
TCP, on the other hand, offers several additional services to applications. First
and foremost, it provides reliable data transfer. Using flow control, sequence numbers, acknowledgments, and timers (techniques we’ll explore in detail in this chapter), TCP ensures that data is delivered from sending process to receiving process,
correctly and in order. TCP thus converts IP’s unreliable service between end systems into a reliable data transport service between processes. TCP also provides
congestion control. Congestion control is not so much a service provided to the
invoking application as it is a service for the Internet as a whole, a service for the
general good. Loosely speaking, TCP congestion control prevents any one TCP connection from swamping the links and routers between communicating hosts with an
excessive amount of traffic. TCP strives to give each connection traversing a congested link an equal share of the link bandwidth. This is done by regulating the rate
at which the sending sides of TCP connections can send traffic into the network.
UDP traffic, on the other hand, is unregulated. An application using UDP transport
can send at any rate it pleases, for as long as it pleases.
A protocol that provides reliable data transfer and congestion control is necessarily complex. We’ll need several sections to cover the principles of reliable data
transfer and congestion control, and additional sections to cover the TCP protocol
itself. These topics are investigated in Sections 3.4 through 3.8. The approach taken
in this chapter is to alternate between basic principles and the TCP protocol. For
example, we’ll first discuss reliable data transfer in a general setting and then discuss how TCP specifically provides reliable data transfer. Similarly, we’ll first
3.2
•
MULTIPLEXING AND DEMULTIPLEXING
discuss congestion control in a general setting and then discuss how TCP performs
congestion control. But before getting into all this good stuff, let’s first look at
transport-layer multiplexing and demultiplexing.
3.2 Multiplexing and Demultiplexing
In this section, we discuss transport-layer multiplexing and demultiplexing, that is,
extending the host-to-host delivery service provided by the network layer to a
process-to-process delivery service for applications running on the hosts. In order to
keep the discussion concrete, we’ll discuss this basic transport-layer service in the
context of the Internet. We emphasize, however, that a multiplexing/demultiplexing
service is needed for all computer networks.
At the destination host, the transport layer receives segments from the network
layer just below. The transport layer has the responsibility of delivering the data in
these segments to the appropriate application process running in the host. Let’s take
a look at an example. Suppose you are sitting in front of your computer, and you are
downloading Web pages while running one FTP session and two Telnet sessions.
You therefore have four network application processes running—two Telnet
processes, one FTP process, and one HTTP process. When the transport layer in
your computer receives data from the network layer below, it needs to direct the
received data to one of these four processes. Let’s now examine how this is done.
First recall from Section 2.7 that a process (as part of a network application) can
have one or more sockets, doors through which data passes from the network to the
process and through which data passes from the process to the network. Thus, as
shown in Figure 3.2, the transport layer in the receiving host does not actually
deliver data directly to a process, but instead to an intermediary socket. Because at
any given time there can be more than one socket in the receiving host, each socket
has a unique identifier. The format of the identifier depends on whether the socket is
a UDP or a TCP socket, as we’ll discuss shortly.
Now let’s consider how a receiving host directs an incoming transport-layer segment to the appropriate socket. Each transport-layer segment has a set of fields in the
segment for this purpose. At the receiving end, the transport layer examines these
fields to identify the receiving socket and then directs the segment to that socket. This
job of delivering the data in a transport-layer segment to the correct socket is called
demultiplexing. The job of gathering data chunks at the source host from different
sockets, encapsulating each data chunk with header information (that will later be
used in demultiplexing) to create segments, and passing the segments to the network
layer is called multiplexing. Note that the transport layer in the middle host in Figure 3.2 must demultiplex segments arriving from the network layer below to either
process P1 or P2 above; this is done by directing the arriving segment’s data to the
corresponding process’s socket. The transport layer in the middle host must also
191
192
CHAPTER 3
•
Application
TRANSPORT LAYER
P1
P3
Application
P2
P4
Application
Transport
Transport
Transport
Network
Network
Network
Data link
Data link
Data link
Physical
Physical
Physical
Key:
Process
Socket
Figure 3.2
Transport-layer multiplexing and demultiplexing
gather outgoing data from these sockets, form transport-layer segments, and pass
these segments down to the network layer. Although we have introduced multiplexing and demultiplexing in the context of the Internet transport protocols, it’s important to realize that they are concerns whenever a single protocol at one layer (at the
transport layer or elsewhere) is used by multiple protocols at the next higher layer.
To illustrate the demultiplexing job, recall the household analogy in the previous section. Each of the kids is identified by his or her name. When Bill receives a
batch of mail from the mail carrier, he performs a demultiplexing operation by
observing to whom the letters are addressed and then hand delivering the mail to his
brothers and sisters. Ann performs a multiplexing operation when she collects letters from her brothers and sisters and gives the collected mail to the mail person.
Now that we understand the roles of transport-layer multiplexing and demultiplexing, let us examine how it is actually done in a host. From the discussion above,
we know that transport-layer multiplexing requires (1) that sockets have unique
identifiers, and (2) that each segment have special fields that indicate the socket to
which the segment is to be delivered. These special fields, illustrated in Figure 3.3,
are the source port number field and the destination port number field. (The
UDP and TCP segments have other fields as well, as discussed in the subsequent
sections of this chapter.) Each port number is a 16-bit number, ranging from 0 to
65535. The port numbers ranging from 0 to 1023 are called well-known port numbers and are restricted, which means that they are reserved for use by well-known
application protocols such as HTTP (which uses port number 80) and FTP (which
uses port number 21). The list of well-known port numbers is given in RFC 1700
and is updated at http://www.iana.org [RFC 3232]. When we develop a new
3.2
•
MULTIPLEXING AND DEMULTIPLEXING
32 bits
Source port #
Dest. port #
Other header fields
Application
data
(message)
Figure 3.3
Source and destination port-number fields in a transport-layer
segment
application (such as the simple application developed in Section 2.7), we must
assign the application a port number.
It should now be clear how the transport layer could implement the demultiplexing service: Each socket in the host could be assigned a port number, and when a segment arrives at the host, the transport layer examines the destination port number in
the segment and directs the segment to the corresponding socket. The segment’s data
then passes through the socket into the attached process. As we’ll see, this is basically how UDP does it. However, we’ll also see that multiplexing/demultiplexing in
TCP is yet more subtle.
Connectionless Multiplexing and Demultiplexing
Recall from Section 2.7.1 that the Python program running in a host can create a
UDP socket with the line
clientSocket = socket(socket.AF_INET, socket.SOCK_DGRAM)
When a UDP socket is created in this manner, the transport layer automatically
assigns a port number to the socket. In particular, the transport layer assigns a port
number in the range 1024 to 65535 that is currently not being used by any other UDP
port in the host. Alternatively, we can add a line into our Python program after we
create the socket to associate a specific port number (say, 19157) to this UDP socket
via the socket bind() method:
clientSocket.bind((‘’, 19157))
If the application developer writing the code were implementing the server side of a
“well-known protocol,” then the developer would have to assign the corresponding
193
194
CHAPTER 3
•
TRANSPORT LAYER
well-known port number. Typically, the client side of the application lets the transport layer automatically (and transparently) assign the port number, whereas the
server side of the application assigns a specific port number.
With port numbers assigned to UDP sockets, we can now precisely describe
UDP multiplexing/demultiplexing. Suppose a process in Host A, with UDP port
19157, wants to send a chunk of application data to a process with UDP port 46428
in Host B. The transport layer in Host A creates a transport-layer segment that
includes the application data, the source port number (19157), the destination port
number (46428), and two other values (which will be discussed later, but are unimportant for the current discussion). The transport layer then passes the resulting segment to the network layer. The network layer encapsulates the segment in an IP
datagram and makes a best-effort attempt to deliver the segment to the receiving host.
If the segment arrives at the receiving Host B, the transport layer at the receiving
host examines the destination port number in the segment (46428) and delivers the
segment to its socket identified by port 46428. Note that Host B could be running
multiple processes, each with its own UDP socket and associated port number. As
UDP segments arrive from the network, Host B directs (demultiplexes) each segment
to the appropriate socket by examining the segment’s destination port number.
It is important to note that a UDP socket is fully identified by a two-tuple consisting of a destination IP address and a destination port number. As a consequence, if two
UDP segments have different source IP addresses and/or source port numbers, but have
the same destination IP address and destination port number, then the two segments
will be directed to the same destination process via the same destination socket.
You may be wondering now, what is the purpose of the source port number? As
shown in Figure 3.4, in the A-to-B segment the source port number serves as part of
a “return address”—when B wants to send a segment back to A, the destination port
in the B-to-A segment will take its value from the source port value of the A-to-B
segment. (The complete return address is A’s IP address and the source port number.) As an example, recall the UDP server program studied in Section 2.7. In
UDPServer.py, the server uses the recvfrom() method to extract the clientside (source) port number from the segment it receives from the client; it then sends
a new segment to the client, with the extracted source port number serving as the
destination port number in this new segment.
Connection-Oriented Multiplexing and Demultiplexing
In order to understand TCP demultiplexing, we have to take a close look at TCP
sockets and TCP connection establishment. One subtle difference between a TCP
socket and a UDP socket is that a TCP socket is identified by a four-tuple: (source
IP address, source port number, destination IP address, destination port number).
Thus, when a TCP segment arrives from the network to a host, the host uses all four
values to direct (demultiplex) the segment to the appropriate socket. In particular,
and in contrast with UDP, two arriving TCP segments with different source IP
3.2
•
MULTIPLEXING AND DEMULTIPLEXING
Client process
Socket
Host A
Server B
source port:
19157
dest. port:
46428
source port:
46428
Figure 3.4
dest. port:
19157
The inversion of source and destination port numbers
addresses or source port numbers will (with the exception of a TCP segment carrying the original connection-establishment request) be directed to two different sockets. To gain further insight, let’s reconsider the TCP client-server programming
example in Section 2.7.2:
• The TCP server application has a “welcoming socket,” that waits for connectionestablishment requests from TCP clients (see Figure 2.29) on port number 12000.
• The TCP client creates a socket and sends a connection establishment request
segment with the lines:
clientSocket = socket(AF_INET, SOCK_STREAM)
clientSocket.connect((serverName,12000))
• A connection-establishment request is nothing more than a TCP segment with destination port number 12000 and a special connection-establishment bit set in the TCP
header (discussed in Section 3.5). The segment also includes a source port number
that was chosen by the client.
• When the host operating system of the computer running the server process
receives the incoming connection-request segment with destination port 12000,
it locates the server process that is waiting to accept a connection on port number 12000. The server process then creates a new socket:
connectionSocket, addr = serverSocket.accept()
195
196
CHAPTER 3
•
TRANSPORT LAYER
• Also, the transport layer at the server notes the following four values in the connection-request segment: (1) the source port number in the segment, (2) the IP
address of the source host, (3) the destination port number in the segment, and
(4) its own IP address. The newly created connection socket is identified by these
four values; all subsequently arriving segments whose source port, source IP
address, destination port, and destination IP address match these four values will
be demultiplexed to this socket. With the TCP connection now in place, the client
and server can now send data to each other.
The server host may support many simultaneous TCP connection sockets, with
each socket attached to a process, and with each socket identified by its own fourtuple. When a TCP segment arrives at the host, all four fields (source IP address,
source port, destination IP address, destination port) are used to direct (demultiplex)
the segment to the appropriate socket.
FOCUS ON SECURITY
PORT SCANNING
We’ve seen that a server process waits patiently on an open port for contact by a
remote client. Some ports are reserved for well-known applications (e.g., Web, FTP,
DNS, and SMTP servers); other ports are used by convention by popular applications
(e.g., the Microsoft 2000 SQL server listens for requests on UDP port 1434). Thus, if
we determine that a port is open on a host, we may be able to map that port to a
specific application running on the host. This is very useful for system administrators,
who are often interested in knowing which network applications are running on the
hosts in their networks. But attackers, in order to “case the joint,” also want to know
which ports are open on target hosts. If a host is found to be running an application
with a known security flaw (e.g., a SQL server listening on port 1434 was subject to
a buffer overflow, allowing a remote user to execute arbitrary code on the vulnerable
host, a flaw exploited by the Slammer worm [CERT 2003–04]), then that host is ripe
for attack.
Determining which applications are listening on which ports is a relatively easy
task. Indeed there are a number of public domain programs, called port scanners,
that do just that. Perhaps the most widely used of these is nmap, freely available at
http://nmap.org and included in most Linux distributions. For TCP, nmap sequentially
scans ports, looking for ports that are accepting TCP connections. For UDP, nmap
again sequentially scans ports, looking for UDP ports that respond to transmitted
UDP segments. In both cases, nmap returns a list of open, closed, or unreachable
ports. A host running nmap can attempt to scan any target host anywhere in the
Internet. We’ll revisit nmap in Section 3.5.6, when we discuss TCP connection
management.
3.2
•
MULTIPLEXING AND DEMULTIPLEXING
Web client
host C
Web
server B
source port:
7532
dest. port:
80
source port:
26145
dest. port:
80
source IP:
C
dest. IP:
B
source IP:
C
dest. IP:
B
Web client
host A
Figure 3.5
source port:
26145
dest. port:
80
source IP:
A
dest. IP:
B
Two clients, using the same destination port number (80) to
communicate with the same Web server application
The situation is illustrated in Figure 3.5, in which Host C initiates two HTTP sessions to server B, and Host A initiates one HTTP session to B. Hosts A and C and server
B each have their own unique IP address—A, C, and B, respectively. Host C assigns
two different source port numbers (26145 and 7532) to its two HTTP connections.
Because Host A is choosing source port numbers independently of C, it might also
assign a source port of 26145 to its HTTP connection. But this is not a problem—server
B will still be able to correctly demultiplex the two connections having the same source
port number, since the two connections have different source IP addresses.
Web Servers and TCP
Before closing this discussion, it’s instructive to say a few additional words about
Web servers and how they use port numbers. Consider a host running a Web server,
such as an Apache Web server, on port 80. When clients (for example, browsers)
send segments to the server, all segments will have destination port 80. In particular, both the initial connection-establishment segments and the segments carrying
HTTP request messages will have destination port 80. As we have just described,
197
Per-connection
HTTP
processes
Transportlayer
demultiplexing
198
CHAPTER 3
•
TRANSPORT LAYER
the server distinguishes the segments from the different clients using source IP
addresses and source port numbers.
Figure 3.5 shows a Web server that spawns a new process for each connection.
As shown in Figure 3.5, each of these processes has its own connection socket
through which HTTP requests arrive and HTTP responses are sent. We mention,
however, that there is not always a one-to-one correspondence between connection
sockets and processes. In fact, today’s high-performing Web servers often use only
one process, and create a new thread with a new connection socket for each new
client connection. (A thread can be viewed as a lightweight subprocess.) If you did
the first programming assignment in Chapter 2, you built a Web server that does just
this. For such a server, at any given time there may be many connection sockets
(with different identifiers) attached to the same process.
If the client and server are using persistent HTTP, then throughout the duration
of the persistent connection the client and server exchange HTTP messages via the
same server socket. However, if the client and server use non-persistent HTTP, then
a new TCP connection is created and closed for every request/response, and hence
a new socket is created and later closed for every request/response. This frequent
creating and closing of sockets can severely impact the performance of a busy Web
server (although a number of operating system tricks can be used to mitigate
the problem). Readers interested in the operating system issues surrounding persistent and non-persistent HTTP are encouraged to see [Nielsen 1997; Nahum
2002].
Now that we’ve discussed transport-layer multiplexing and demultiplexing,
let’s move on and discuss one of the Internet’s transport protocols, UDP. In the next
section we’ll see that UDP adds little more to the network-layer protocol than a multiplexing/demultiplexing service.
3.3 Connectionless Transport: UDP
In this section, we’ll take a close look at UDP, how it works, and what it does.
We encourage you to refer back to Section 2.1, which includes an overview of
the UDP service model, and to Section 2.7.1, which discusses socket programming using UDP.
To motivate our discussion about UDP, suppose you were interested in designing a no-frills, bare-bones transport protocol. How might you go about doing this?
You might first consider using a vacuous transport protocol. In particular, on the
sending side, you might consider taking the messages from the application process
and passing them directly to the network layer; and on the receiving side, you might
consider taking the messages arriving from the network layer and passing them
directly to the application process. But as we learned in the previous section, we
have to do a little more than nothing! At the very least, the transport layer has to
3.3
•
CONNECTIONLESS TRANSPORT: UDP
provide a multiplexing/demultiplexing service in order to pass data between the
network layer and the correct application-level process.
UDP, defined in [RFC 768], does just about as little as a transport protocol can
do. Aside from the multiplexing/demultiplexing function and some light error
checking, it adds nothing to IP. In fact, if the application developer chooses UDP
instead of TCP, then the application is almost directly talking with IP. UDP takes
messages from the application process, attaches source and destination port number
fields for the multiplexing/demultiplexing service, adds two other small fields, and
passes the resulting segment to the network layer. The network layer encapsulates
the transport-layer segment into an IP datagram and then makes a best-effort attempt
to deliver the segment to the receiving host. If the segment arrives at the receiving
host, UDP uses the destination port number to deliver the segment’s data to the correct application process. Note that with UDP there is no handshaking between sending and receiving transport-layer entities before sending a segment. For this reason,
UDP is said to be connectionless.
DNS is an example of an application-layer protocol that typically uses UDP.
When the DNS application in a host wants to make a query, it constructs a DNS
query message and passes the message to UDP. Without performing any handshaking with the UDP entity running on the destination end system, the host-side UDP
adds header fields to the message and passes the resulting segment to the network
layer. The network layer encapsulates the UDP segment into a datagram and sends
the datagram to a name server. The DNS application at the querying host then waits
for a reply to its query. If it doesn’t receive a reply (possibly because the underlying
network lost the query or the reply), either it tries sending the query to another name
server, or it informs the invoking application that it can’t get a reply.
Now you might be wondering why an application developer would ever choose
to build an application over UDP rather than over TCP. Isn’t TCP always preferable,
since TCP provides a reliable data transfer service, while UDP does not? The answer
is no, as many applications are better suited for UDP for the following reasons:
• Finer application-level control over what data is sent, and when. Under UDP, as
soon as an application process passes data to UDP, UDP will package the data
inside a UDP segment and immediately pass the segment to the network layer.
TCP, on the other hand, has a congestion-control mechanism that throttles the
transport-layer TCP sender when one or more links between the source and destination hosts become excessively congested. TCP will also continue to resend a
segment until the receipt of the segment has been acknowledged by the destination, regardless of how long reliable delivery takes. Since real-time applications
often require a minimum sending rate, do not want to overly delay segment
transmission, and can tolerate some data loss, TCP’s service model is not particularly well matched to these applications’ needs. As discussed below, these applications can use UDP and implement, as part of the application, any additional
functionality that is needed beyond UDP’s no-frills segment-delivery service.
199
200
CHAPTER 3
•
TRANSPORT LAYER
• No connection establishment. As we’ll discuss later, TCP uses a three-way handshake before it starts to transfer data. UDP just blasts away without any formal preliminaries. Thus UDP does not introduce any delay to establish a connection. This
is probably the principal reason why DNS runs over UDP rather than TCP—DNS
would be much slower if it ran over TCP. HTTP uses TCP rather than UDP, since
reliability is critical for Web pages with text. But, as we briefly discussed in Section 2.2, the TCP connection-establishment delay in HTTP is an important contributor to the delays associated with downloading Web documents.
• No connection state. TCP maintains connection state in the end systems. This
connection state includes receive and send buffers, congestion-control parameters, and sequence and acknowledgment number parameters. We will see in Section 3.5 that this state information is needed to implement TCP’s reliable data
transfer service and to provide congestion control. UDP, on the other hand, does
not maintain connection state and does not track any of these parameters. For this
reason, a server devoted to a particular application can typically support many
more active clients when the application runs over UDP rather than TCP.
• Small packet header overhead. The TCP segment has 20 bytes of header overhead in every segment, whereas UDP has only 8 bytes of overhead.
Figure 3.6 lists popular Internet applications and the transport protocols that they
use. As we expect, e-mail, remote terminal access, the Web, and file transfer run over
TCP—all these applications need the reliable data transfer service of TCP. Nevertheless, many important applications run over UDP rather than TCP. UDP is used for RIP
routing table updates (see Section 4.6.1). Since RIP updates are sent periodically (typically every five minutes), lost updates will be replaced by more recent updates, thus
making the lost, out-of-date update useless. UDP is also used to carry network management (SNMP; see Chapter 9) data. UDP is preferred to TCP in this case, since network
management applications must often run when the network is in a stressed state—precisely when reliable, congestion-controlled data transfer is difficult to achieve. Also,
as we mentioned earlier, DNS runs over UDP, thereby avoiding TCP’s connectionestablishment delays.
As shown in Figure 3.6, both UDP and TCP are used today with multimedia
applications, such as Internet phone, real-time video conferencing, and streaming of
stored audio and video. We’ll take a close look at these applications in Chapter 7. We
just mention now that all of these applications can tolerate a small amount of packet
loss, so that reliable data transfer is not absolutely critical for the application’s success. Furthermore, real-time applications, like Internet phone and video conferencing, react very poorly to TCP’s congestion control. For these reasons, developers of
multimedia applications may choose to run their applications over UDP instead of
TCP. However, TCP is increasingly being used for streaming media transport. For
example, [Sripanidkulchai 2004] found that nearly 75% of on-demand and live
streaming used TCP. When packet loss rates are low, and with some organizations
3.3
•
CONNECTIONLESS TRANSPORT: UDP
Application
Application-Layer
Protocol
Underlying Transport
Protocol
Electronic mail
SMTP
TCP
Remote terminal access
Telnet
TCP
Web
HTTP
TCP
File transfer
FTP
TCP
Remote file server
NFS
Typically UDP
Streaming multimedia
typically proprietary
UDP or TCP
Internet telephony
typically proprietary
UDP or TCP
Network management
SNMP
Typically UDP
Routing protocol
RIP
Typically UDP
Name translation
DNS
Typically UDP
Figure 3.6
Popular Internet applications and their underlying transport
protocols
blocking UDP traffic for security reasons (see Chapter 8), TCP becomes an increasingly attractive protocol for streaming media transport.
Although commonly done today, running multimedia applications over UDP is
controversial. As we mentioned above, UDP has no congestion control. But congestion control is needed to prevent the network from entering a congested state in
which very little useful work is done. If everyone were to start streaming high-bitrate video without using any congestion control, there would be so much packet
overflow at routers that very few UDP packets would successfully traverse the
source-to-destination path. Moreover, the high loss rates induced by the uncontrolled UDP senders would cause the TCP senders (which, as we’ll see, do decrease
their sending rates in the face of congestion) to dramatically decrease their rates.
Thus, the lack of congestion control in UDP can result in high loss rates between a
UDP sender and receiver, and the crowding out of TCP sessions—a potentially serious problem [Floyd 1999]. Many researchers have proposed new mechanisms to
force all sources, including UDP sources, to perform adaptive congestion control
[Mahdavi 1997; Floyd 2000; Kohler 2006: RFC 4340].
Before discussing the UDP segment structure, we mention that it is possible for
an application to have reliable data transfer when using UDP. This can be done if reliability is built into the application itself (for example, by adding acknowledgment
and retransmission mechanisms, such as those we’ll study in the next section). But
this is a nontrivial task that would keep an application developer busy debugging for
201
202
CHAPTER 3
•
TRANSPORT LAYER
a long time. Nevertheless, building reliability directly into the application allows the
application to “have its cake and eat it too.” That is, application processes can communicate reliably without being subjected to the transmission-rate constraints
imposed by TCP’s congestion-control mechanism.
3.3.1 UDP Segment Structure
The UDP segment structure, shown in Figure 3.7, is defined in RFC 768. The application data occupies the data field of the UDP segment. For example, for DNS, the data
field contains either a query message or a response message. For a streaming audio
application, audio samples fill the data field. The UDP header has only four fields,
each consisting of two bytes. As discussed in the previous section, the port numbers
allow the destination host to pass the application data to the correct process running
on the destination end system (that is, to perform the demultiplexing function). The
length field specifies the number of bytes in the UDP segment (header plus data). An
explicit length value is needed since the size of the data field may differ from one UDP
segment to the next. The checksum is used by the receiving host to check whether
errors have been introduced into the segment. In truth, the checksum is also calculated
over a few of the fields in the IP header in addition to the UDP segment. But we ignore
this detail in order to see the forest through the trees. We’ll discuss the checksum calculation below. Basic principles of error detection are described in Section 5.2. The
length field specifies the length of the UDP segment, including the header, in bytes.
3.3.2 UDP Checksum
The UDP checksum provides for error detection. That is, the checksum is used to
determine whether bits within the UDP segment have been altered (for example, by
noise in the links or while stored in a router) as it moved from source to destination.
UDP at the sender side performs the 1s complement of the sum of all the 16-bit
words in the segment, with any overflow encountered during the sum being
32 bits
Source port #
Dest. port #
Length
Checksum
Application
data
(message)
Figure 3.7
UDP segment structure
3.3
•
CONNECTIONLESS TRANSPORT: UDP
wrapped around. This result is put in the checksum field of the UDP segment. Here
we give a simple example of the checksum calculation. You can find details about
efficient implementation of the calculation in RFC 1071 and performance over real
data in [Stone 1998; Stone 2000]. As an example, suppose that we have the following three 16-bit words:
0110011001100000
0101010101010101
1000111100001100
The sum of first two of these 16-bit words is
0110011001100000
0101010101010101
1011101110110101
Adding the third word to the above sum gives
1011101110110101
1000111100001100
0100101011000010
Note that this last addition had overflow, which was wrapped around. The 1s complement is obtained by converting all the 0s to 1s and converting all the 1s to 0s.
Thus the 1s complement of the sum 0100101011000010 is 1011010100111101,
which becomes the checksum. At the receiver, all four 16-bit words are added,
including the checksum. If no errors are introduced into the packet, then clearly the
sum at the receiver will be 1111111111111111. If one of the bits is a 0, then we know
that errors have been introduced into the packet.
You may wonder why UDP provides a checksum in the first place, as many linklayer protocols (including the popular Ethernet protocol) also provide error checking.
The reason is that there is no guarantee that all the links between source and destination
provide error checking; that is, one of the links may use a link-layer protocol that does
not provide error checking. Furthermore, even if segments are correctly transferred
across a link, it’s possible that bit errors could be introduced when a segment is stored
in a router’s memory. Given that neither link-by-link reliability nor in-memory error
detection is guaranteed, UDP must provide error detection at the transport layer, on an
end-end basis, if the end-end data transfer service is to provide error detection. This is
an example of the celebrated end-end principle in system design [Saltzer 1984], which
states that since certain functionality (error detection, in this case) must be implemented
on an end-end basis: “functions placed at the lower levels may be redundant or of little
value when compared to the cost of providing them at the higher level.”
Because IP is supposed to run over just about any layer-2 protocol, it is useful
for the transport layer to provide error checking as a safety measure. Although UDP
203
204
CHAPTER 3
•
TRANSPORT LAYER
provides error checking, it does not do anything to recover from an error. Some
implementations of UDP simply discard the damaged segment; others pass the damaged segment to the application with a warning.
That wraps up our discussion of UDP. We will soon see that TCP offers reliable
data transfer to its applications as well as other services that UDP doesn’t offer. Naturally, TCP is also more complex than UDP. Before discussing TCP, however, it will be
useful to step back and first discuss the underlying principles of reliable data transfer.
3.4 Principles of Reliable Data Transfer
In this section, we consider the problem of reliable data transfer in a general context. This is appropriate since the problem of implementing reliable data transfer
occurs not only at the transport layer, but also at the link layer and the application
layer as well. The general problem is thus of central importance to networking.
Indeed, if one had to identify a “top-ten” list of fundamentally important problems
in all of networking, this would be a candidate to lead the list. In the next section
we’ll examine TCP and show, in particular, that TCP exploits many of the principles
that we are about to describe.
Figure 3.8 illustrates the framework for our study of reliable data transfer. The
service abstraction provided to the upper-layer entities is that of a reliable channel
through which data can be transferred. With a reliable channel, no transferred data
bits are corrupted (flipped from 0 to 1, or vice versa) or lost, and all are delivered in
the order in which they were sent. This is precisely the service model offered by
TCP to the Internet applications that invoke it.
It is the responsibility of a reliable data transfer protocol to implement this
service abstraction. This task is made difficult by the fact that the layer below the
reliable data transfer protocol may be unreliable. For example, TCP is a reliable data
transfer protocol that is implemented on top of an unreliable (IP) end-to-end network layer. More generally, the layer beneath the two reliably communicating end
points might consist of a single physical link (as in the case of a link-level data
transfer protocol) or a global internetwork (as in the case of a transport-level protocol). For our purposes, however, we can view this lower layer simply as an unreliable point-to-point channel.
In this section, we will incrementally develop the sender and receiver sides of a
reliable data transfer protocol, considering increasingly complex models of the underlying channel. For example, we’ll consider what protocol mechanisms are needed when
the underlying channel can corrupt bits or lose entire packets. One assumption we’ll
adopt throughout our discussion here is that packets will be delivered in the order in
which they were sent, with some packets possibly being lost; that is, the underlying
channel will not reorder packets. Figure 3.8(b) illustrates the interfaces for our data
transfer protocol. The sending side of the data transfer protocol will be invoked from
above by a call to rdt_send(). It will pass the data to be delivered to the upper layer
at the receiving side. (Here rdt stands for reliable data transfer protocol and _send
3.4
Sending
process
Application
layer
•
PRINCIPLES OF RELIABLE DATA TRANSFER
Receiver
process
rdt_send()
Transport
layer
Reliable channel
deliver_data
Reliable data
transfer protocol
(sending side)
Reliable data
transfer protocol
(receiving side)
rdt_rcv()
udt_send()
Network
layer
Unreliable channel
a. Provided service
b. Service implementation
Key:
Data
Figure 3.8
Packet
Reliable data transfer: Service model and service
implementation
indicates that the sending side of rdt is being called. The first step in developing any
protocol is to choose a good name!) On the receiving side, rdt_rcv() will be called
when a packet arrives from the receiving side of the channel. When the rdt protocol
wants to deliver data to the upper layer, it will do so by calling deliver_data(). In
the following we use the terminology “packet” rather than transport-layer “segment.”
Because the theory developed in this section applies to computer networks in general
and not just to the Internet transport layer, the generic term “packet” is perhaps more
appropriate here.
In this section we consider only the case of unidirectional data transfer, that is,
data transfer from the sending to the receiving side. The case of reliable bidirectional
(that is, full-duplex) data transfer is conceptually no more difficult but considerably
more tedious to explain. Although we consider only unidirectional data transfer, it is
important to note that the sending and receiving sides of our protocol will nonetheless
need to transmit packets in both directions, as indicated in Figure 3.8. We will see
shortly that, in addition to exchanging packets containing the data to be transferred, the
205
206
CHAPTER 3
•
TRANSPORT LAYER
sending and receiving sides of rdt will also need to exchange control packets back and
forth. Both the send and receive sides of rdt send packets to the other side by a call to
udt_send() (where udt stands for unreliable data transfer).
3.4.1 Building a Reliable Data Transfer Protocol
We now step through a series of protocols, each one becoming more complex, arriving at a flawless, reliable data transfer protocol.
Reliable Data Transfer over a Perfectly Reliable Channel: rdt1.0
We first consider the simplest case, in which the underlying channel is completely
reliable. The protocol itself, which we’ll call rdt1.0, is trivial. The finite-state
machine (FSM) definitions for the rdt1.0 sender and receiver are shown in
Figure 3.9. The FSM in Figure 3.9(a) defines the operation of the sender, while the
FSM in Figure 3.9(b) defines the operation of the receiver. It is important to note
that there are separate FSMs for the sender and for the receiver. The sender and
receiver FSMs in Figure 3.9 each have just one state. The arrows in the FSM
description indicate the transition of the protocol from one state to another. (Since
each FSM in Figure 3.9 has just one state, a transition is necessarily from the one
state back to itself; we’ll see more complicated state diagrams shortly.) The event
causing the transition is shown above the horizontal line labeling the transition, and
Wait for
call from
above
rdt_send(data)
packet=make_pkt(data)
udt_send(packet)
a. rdt1.0: sending side
Wait for
call from
below
rdt_rcv(packet)
extract(packet,data)
deliver_data(data)
b. rdt1.0: receiving side
Figure 3.9
rdt1.0 – A protocol for a completely reliable channel
3.4
•
PRINCIPLES OF RELIABLE DATA TRANSFER
the actions taken when the event occurs are shown below the horizontal line. When
no action is taken on an event, or no event occurs and an action is taken, we’ll use
the symbol ⌳ below or above the horizontal, respectively, to explicitly denote the
lack of an action or event. The initial state of the FSM is indicated by the dashed
arrow. Although the FSMs in Figure 3.9 have but one state, the FSMs we will see
shortly have multiple states, so it will be important to identify the initial state of
each FSM.
The sending side of rdt simply accepts data from the upper layer via the
rdt_send(data) event, creates a packet containing the data (via the action
make_pkt(data)) and sends the packet into the channel. In practice, the
rdt_send(data) event would result from a procedure call (for example, to
rdt_send()) by the upper-layer application.
On the receiving side, rdt receives a packet from the underlying channel via
the rdt_rcv(packet) event, removes the data from the packet (via the action
extract (packet, data)) and passes the data up to the upper layer (via the
action deliver_data(data)). In practice, the rdt_rcv(packet) event
would result from a procedure call (for example, to rdt_rcv()) from the lowerlayer protocol.
In this simple protocol, there is no difference between a unit of data and a
packet. Also, all packet flow is from the sender to receiver; with a perfectly reliable
channel there is no need for the receiver side to provide any feedback to the sender
since nothing can go wrong! Note that we have also assumed that the receiver is able
to receive data as fast as the sender happens to send data. Thus, there is no need for
the receiver to ask the sender to slow down!
Reliable Data Transfer over a Channel with Bit Errors: rdt2.0
A more realistic model of the underlying channel is one in which bits in a packet
may be corrupted. Such bit errors typically occur in the physical components of a
network as a packet is transmitted, propagates, or is buffered. We’ll continue to
assume for the moment that all transmitted packets are received (although their bits
may be corrupted) in the order in which they were sent.
Before developing a protocol for reliably communicating over such a channel,
first consider how people might deal with such a situation. Consider how you yourself might dictate a long message over the phone. In a typical scenario, the message
taker might say “OK” after each sentence has been heard, understood, and recorded.
If the message taker hears a garbled sentence, you’re asked to repeat the garbled
sentence. This message-dictation protocol uses both positive acknowledgments
(“OK”) and negative acknowledgments (“Please repeat that.”). These control messages allow the receiver to let the sender know what has been received correctly, and
what has been received in error and thus requires repeating. In a computer network
setting, reliable data transfer protocols based on such retransmission are known as
ARQ (Automatic Repeat reQuest) protocols.
207
208
CHAPTER 3
•
TRANSPORT LAYER
Fundamentally, three additional protocol capabilities are required in ARQ
protocols to handle the presence of bit errors:
• Error detection. First, a mechanism is needed to allow the receiver to detect
when bit errors have occurred. Recall from the previous section that UDP uses
the Internet checksum field for exactly this purpose. In Chapter 5 we’ll examine error-detection and -correction techniques in greater detail; these techniques allow the receiver to detect and possibly correct packet bit errors. For
now, we need only know that these techniques require that extra bits (beyond
the bits of original data to be transferred) be sent from the sender to the
receiver; these bits will be gathered into the packet checksum field of the
rdt2.0 data packet.
• Receiver feedback. Since the sender and receiver are typically executing on different end systems, possibly separated by thousands of miles, the only way for the
sender to learn of the receiver’s view of the world (in this case, whether or not a
packet was received correctly) is for the receiver to provide explicit feedback to the
sender. The positive (ACK) and negative (NAK) acknowledgment replies in the
message-dictation scenario are examples of such feedback. Our rdt2.0 protocol
will similarly send ACK and NAK packets back from the receiver to the sender. In
principle, these packets need only be one bit long; for example, a 0 value could indicate a NAK and a value of 1 could indicate an ACK.
• Retransmission. A packet that is received in error at the receiver will be retransmitted by the sender.
Figure 3.10 shows the FSM representation of rdt2.0, a data transfer protocol
employing error detection, positive acknowledgments, and negative acknowledgments.
The send side of rdt2.0 has two states. In the leftmost state, the send-side protocol is waiting for data to be passed down from the upper layer. When the
rdt_send(data) event occurs, the sender will create a packet (sndpkt)
containing the data to be sent, along with a packet checksum (for example, as discussed
in Section 3.3.2 for the case of a UDP segment), and then send the packet via the
udt_send(sndpkt) operation. In the rightmost state, the sender protocol is waiting for an ACK or a NAK packet from the receiver. If an ACK packet is received (the
notation rdt_rcv(rcvpkt) && isACK (rcvpkt) in Figure 3.10 corresponds
to this event), the sender knows that the most recently transmitted packet has been
received correctly and thus the protocol returns to the state of waiting for data from the
upper layer. If a NAK is received, the protocol retransmits the last packet and waits for
an ACK or NAK to be returned by the receiver in response to the retransmitted data
packet. It is important to note that when the sender is in the wait-for-ACK-or-NAK
state, it cannot get more data from the upper layer; that is, the rdt_send() event can
not occur; that will happen only after the sender receives an ACK and leaves this state.
Thus, the sender will not send a new piece of data until it is sure that the receiver has
3.4
•
PRINCIPLES OF RELIABLE DATA TRANSFER
rdt_send(data)
sndpkt=make_pkt(data,checksum)
udt_send(sndpkt)
Wait for
call from
above
Wait for
ACK or
NAK
rdt_rcv(rcvpkt) && isNAK(rcvpkt)
udt_send(sndpkt)
rdt_rcv(rcvpkt) && isACK(rcvpkt)
Λ
a. rdt2.0: sending side
rdt_rcv(rcvpkt) && corrupt(rcvpkt)
sndpkt=make_pkt(NAK)
udt_send(sndpkt)
Wait for
call from
below
rdt_rcv(rcvpkt) && notcorrupt(rcvpkt)
extract(rcvpkt,data)
deliver_data(data)
sndpkt=make_pkt(ACK)
udt_send(sndpkt)
b. rdt2.0: receiving side
Figure 3.10
rdt2.0–A protocol for a channel with bit errors
correctly received the current packet. Because of this behavior, protocols such as
rdt2.0 are known as stop-and-wait protocols.
The receiver-side FSM for rdt2.0 still has a single state. On packet arrival,
the receiver replies with either an ACK or a NAK, depending on whether or not the
received packet is corrupted. In Figure 3.10, the notation rdt_rcv(rcvpkt) &&
corrupt(rcvpkt) corresponds to the event in which a packet is received and is
found to be in error.
Protocol rdt2.0 may look as if it works but, unfortunately, it has a fatal
flaw. In particular, we haven’t accounted for the possibility that the ACK or NAK
packet could be corrupted! (Before proceeding on, you should think about how this
209
210
CHAPTER 3
•
TRANSPORT LAYER
problem may be fixed.) Unfortunately, our slight oversight is not as innocuous as it
may seem. Minimally, we will need to add checksum bits to ACK/NAK packets in
order to detect such errors. The more difficult question is how the protocol should
recover from errors in ACK or NAK packets. The difficulty here is that if an ACK
or NAK is corrupted, the sender has no way of knowing whether or not the receiver
has correctly received the last piece of transmitted data.
Consider three possibilities for handling corrupted ACKs or NAKs:
• For the first possibility, consider what a human might do in the messagedictation scenario. If the speaker didn’t understand the “OK” or “Please repeat
that” reply from the receiver, the speaker would probably ask, “What did you
say?” (thus introducing a new type of sender-to-receiver packet to our protocol).
The receiver would then repeat the reply. But what if the speaker’s “What did
you say?” is corrupted? The receiver, having no idea whether the garbled sentence was part of the dictation or a request to repeat the last reply, would probably then respond with “What did you say?” And then, of course, that response
might be garbled. Clearly, we’re heading down a difficult path.
• A second alternative is to add enough checksum bits to allow the sender not only
to detect, but also to recover from, bit errors. This solves the immediate problem
for a channel that can corrupt packets but not lose them.
• A third approach is for the sender simply to resend the current data packet when
it receives a garbled ACK or NAK packet. This approach, however, introduces
duplicate packets into the sender-to-receiver channel. The fundamental difficulty with duplicate packets is that the receiver doesn’t know whether the ACK
or NAK it last sent was received correctly at the sender. Thus, it cannot know a
priori whether an arriving packet contains new data or is a retransmission!
A simple solution to this new problem (and one adopted in almost all existing
data transfer protocols, including TCP) is to add a new field to the data packet and
have the sender number its data packets by putting a sequence number into this
field. The receiver then need only check this sequence number to determine whether
or not the received packet is a retransmission. For this simple case of a stop-andwait protocol, a 1-bit sequence number will suffice, since it will allow the receiver
to know whether the sender is resending the previously transmitted packet (the
sequence number of the received packet has the same sequence number as the most
recently received packet) or a new packet (the sequence number changes, moving
“forward” in modulo-2 arithmetic). Since we are currently assuming a channel that
does not lose packets, ACK and NAK packets do not themselves need to indicate
the sequence number of the packet they are acknowledging. The sender knows that
a received ACK or NAK packet (whether garbled or not) was generated in response
to its most recently transmitted data packet.
3.4
•
PRINCIPLES OF RELIABLE DATA TRANSFER
Figures 3.11 and 3.12 show the FSM description for rdt2.1, our fixed version
of rdt2.0. The rdt2.1 sender and receiver FSMs each now have twice as many
states as before. This is because the protocol state must now reflect whether the packet
currently being sent (by the sender) or expected (at the receiver) should have a
sequence number of 0 or 1. Note that the actions in those states where a 0-numbered
packet is being sent or expected are mirror images of those where a 1-numbered
packet is being sent or expected; the only differences have to do with the handling of
the sequence number.
Protocol rdt2.1 uses both positive and negative acknowledgments from the
receiver to the sender. When an out-of-order packet is received, the receiver sends a
positive acknowledgment for the packet it has received. When a corrupted packet is
received, the receiver sends a negative acknowledgment. We can accomplish the
same effect as a NAK if, instead of sending a NAK, we send an ACK for the last
correctly received packet. A sender that receives two ACKs for the same packet (that
is, receives duplicate ACKs) knows that the receiver did not correctly receive the
packet following the packet that is being ACKed twice. Our NAK-free reliable data
rdt_send(data)
sndpkt=make_pkt(0,data,checksum)
udt_send(sndpkt)
rdt_rcv(rcvpkt)&&
(corrupt(rcvpkt)||
isNAK(rcvpkt))
Wait for
call 0 from
above
Wait for
ACK or
NAK 0
rdt_rcv(rcvpkt)
&& notcorrupt(rcvpkt)
&& isACK(rcvpkt)
rdt_rcv(rcvpkt)
&& notcorrupt(rcvpkt)
&& isACK(rcvpkt)
Λ
Λ
Wait for
ACK or
NAK 1
Wait for
call 1 from
above
rdt_rcv(rcvpkt)&&
(corrupt(rcvpkt)||
isNAK(rcvpkt))
udt_send(sndpkt)
rdt_send(data)
sndpkt=make_pkt(1,data,checksum)
udt_send(sndpkt)
Figure 3.11
rdt2.1 sender
udt_send(sndpkt)
211
212
CHAPTER 3
•
TRANSPORT LAYER
rdt_rcv(rcvpkt)&& notcorrupt(rcvpkt)
&& has_seq0(rcvpkt)
extract(rcvpkt,data)
deliver_data(data)
sndpkt=make_pkt(ACK,checksum)
udt_send(sndpkt)
rdt_rcv(rcvpkt)
&& corrupt(rcvpkt)
sndpkt=make_pkt(NAK,checksum)
udt_send(sndpkt)
sndpkt=make_pkt(NAK,checksum)
udt_send(sndpkt)
rdt_rcv(rcvpkt)&& notcorrupt
(rcvpkt)&&has_seq1(rcvpkt)
rdt_rcv(rcvpkt) && corrupt(rcvpkt)
Wait for
0 from
below
Wait for
1 from
below
rdt_rcv(rcvpkt)&& notcorrupt
(rcvpkt)&&has_seq0(rcvpkt)
sndpkt=make_pkt(ACK,checksum)
udt_send(sndpkt)
sndpkt=make_pkt(ACK,checksum)
udt_send(sndpkt)
rdt_rcv(rcvpkt) && notcorrupt(rcvpkt)
&& has_seq1(rcvpkt)
extract(rcvpkt,data)
deliver_data(data)
sndpkt=make_pkt(ACK,checksum)
udt_send(sndpkt)
Figure 3.12
rdt2.1 receiver
transfer protocol for a channel with bit errors is rdt2.2, shown in Figures 3.13 and
3.14. One subtle change between rtdt2.1 and rdt2.2 is that the receiver must
now include the sequence number of the packet being acknowledged by an ACK
message (this is done by including the ACK,0 or ACK,1 argument in make_pkt()
in the receiver FSM), and the sender must now check the sequence number of the
packet being acknowledged by a received ACK message (this is done by including
the 0 or 1 argument in isACK()in the sender FSM).
Reliable Data Transfer over a Lossy Channel with Bit Errors: rdt3.0
Suppose now that in addition to corrupting bits, the underlying channel can lose
packets as well, a not-uncommon event in today’s computer networks (including the
Internet). Two additional concerns must now be addressed by the protocol: how to
detect packet loss and what to do when packet loss occurs. The use of checksumming, sequence numbers, ACK packets, and retransmissions—the techniques
already developed in rdt2.2—will allow us to answer the latter concern. Handling the first concern will require adding a new protocol mechanism.
There are many possible approaches toward dealing with packet loss (several
more of which are explored in the exercises at the end of the chapter). Here, we’ll
put the burden of detecting and recovering from lost packets on the sender. Suppose
3.4
•
PRINCIPLES OF RELIABLE DATA TRANSFER
rdt_send(data)
sndpkt=make_pkt(0,data,checksum)
udt_send(sndpkt)
rdt_rcv(rcvpkt) &&
(corrupt(rcvpkt)||
isACK(rcvpkt,1))
Wait for
call 0 from
above
Wait for
ACK 0
rdt_rcv(rcvpkt)
&& notcorrupt(rcvpkt)
&& isACK(rcvpkt,1)
udt_send(sndpkt)
rdt_rcv(rcvpkt)
&& notcorrupt(rcvpkt)
&& isACK(rcvpkt,0)
Λ
Λ
Wait for
ACK 1
Wait for
call 1 from
above
rdt_rcv(rcvpkt) &&
(corrupt(rcvpkt)||
isACK(rcvpkt,0))
udt_send(sndpkt)
rdt_send(data)
sndpkt=make_pkt(1,data,checksum)
udt_send(sndpkt)
Figure 3.13
rdt2.2 sender
that the sender transmits a data packet and either that packet, or the receiver’s ACK
of that packet, gets lost. In either case, no reply is forthcoming at the sender from
the receiver. If the sender is willing to wait long enough so that it is certain that a
packet has been lost, it can simply retransmit the data packet. You should convince
yourself that this protocol does indeed work.
But how long must the sender wait to be certain that something has been lost?
The sender must clearly wait at least as long as a round-trip delay between the
sender and receiver (which may include buffering at intermediate routers) plus
whatever amount of time is needed to process a packet at the receiver. In many networks, this worst-case maximum delay is very difficult even to estimate, much less
know with certainty. Moreover, the protocol should ideally recover from packet
loss as soon as possible; waiting for a worst-case delay could mean a long wait
until error recovery is initiated. The approach thus adopted in practice is for the
sender to judiciously choose a time value such that packet loss is likely, although
not guaranteed, to have happened. If an ACK is not received within this time, the
packet is retransmitted. Note that if a packet experiences a particularly large delay,
the sender may retransmit the packet even though neither the data packet nor its
ACK have been lost. This introduces the possibility of duplicate data packets in
213
214
CHAPTER 3
•
TRANSPORT LAYER
rdt_rcv(rcvpkt) && notcorrupt(rcvpkt)
&& has_seq0(rcvpkt)
extract(rcvpkt,data)
deliver_data(data)
sndpkt=make_pkt(ACK,0,checksum)
udt_send(sndpkt)
rdt_rcv(rcvpkt) &&
(corrupt(rcvpkt)||
has_seq0(rcvpkt))
sndpkt=make_pkt(ACK,0,che
udt_send(sndpkt)
Wait for
0 from
below
rdt_rcv(rcvpkt) &&
(corrupt(rcvpkt)||
has_seq1(rcvpkt))
Wait for
1 from
below
sndpkt=make_pkt(ACK,1,checksum)
udt_send(sndpkt)
rdt_rcv(rcvpkt) && notcorrupt(rcvpkt)
&& has_seq1(rcvpkt)
extract(rcvpkt,data)
deliver_data(data)
sndpkt=make_pkt(ACK,1,checksum)
udt_send(sndpkt)
Figure 3.14
rdt2.2 receiver
the sender-to-receiver channel. Happily, protocol rdt2.2 already has enough
functionality (that is, sequence numbers) to handle the case of duplicate packets.
From the sender’s viewpoint, retransmission is a panacea. The sender does not
know whether a data packet was lost, an ACK was lost, or if the packet or ACK was
simply overly delayed. In all cases, the action is the same: retransmit. Implementing
a time-based retransmission mechanism requires a countdown timer that can
interrupt the sender after a given amount of time has expired. The sender will thus
need to be able to (1) start the timer each time a packet (either a first-time packet or
a retransmission) is sent, (2) respond to a timer interrupt (taking appropriate
actions), and (3) stop the timer.
Figure 3.15 shows the sender FSM for rdt3.0, a protocol that reliably transfers
data over a channel that can corrupt or lose packets; in the homework problems, you’ll
be asked to provide the receiver FSM for rdt3.0. Figure 3.16 shows how the protocol operates with no lost or delayed packets and how it handles lost data packets. In
Figure 3.16, time moves forward from the top of the diagram toward the bottom of the
diagram; note that a receive time for a packet is necessarily later than the send time
for a packet as a result of transmission and propagation delays. In Figures 3.16(b)–(d),
the send-side brackets indicate the times at which a timer is set and later times out.
Several of the more subtle aspects of this protocol are explored in the exercises at the
end of this chapter. Because packet sequence numbers alternate between 0 and 1, protocol rdt3.0 is sometimes known as the alternating-bit protocol.
3.4
•
PRINCIPLES OF RELIABLE DATA TRANSFER
215
rdt_send(data)
sndpkt=make_pkt(0,data,checksum)
udt_send(sndpkt)
start_timer
rdt_rcv(rcvpkt) &&
(corrupt(rcvpkt)||
isACK(rcvpkt,1))
Λ
rdt_rcv(rcvpkt)
Λ
Wait for
call 0 from
above
timeout
Wait for
ACK 0
rdt_rcv(rcvpkt)
&& notcorrupt(rcvpkt)
&& isACK(rcvpkt,1)
udt_send(sndpkt)
start_timer
rdt_rcv(rcvpkt)
&& notcorrupt(rcvpkt)
&& isACK(rcvpkt,0)
stop_timer
stop_timer
timeout
udt_send(sndpkt)
start_timer
Wait for
ACK 1
Wait for
call 1 from
above
rdt_rcv(rcvpkt)
rdt_rcv(rcvpkt) &&
(corrupt(rcvpkt)||
isACK(rcvpkt,0))
Λ
Figure 3.15
Λ
rdt_send(data)
sndpkt=make_pkt(1,data,checksum)
udt_send(sndpkt)
start_timer
rdt3.0 sender
We have now assembled the key elements of a data transfer protocol. Checksums, sequence numbers, timers, and positive and negative acknowledgment packets each play a crucial and necessary role in the operation of the protocol. We now
have a working reliable data transfer protocol!
3.4.2 Pipelined Reliable Data Transfer Protocols
Protocol rdt3.0 is a functionally correct protocol, but it is unlikely that anyone would
be happy with its performance, particularly in today’s high-speed networks. At the heart
of rdt3.0’s performance problem is the fact that it is a stop-and-wait protocol.
To appreciate the performance impact of this stop-and-wait behavior, consider
an idealized case of two hosts, one located on the West Coast of the United States
and the other located on the East Coast, as shown in Figure 3.17. The speed-of-light
round-trip propagation delay between these two end systems, RTT, is approximately 30 milliseconds. Suppose that they are connected by a channel with a transmission rate, R, of 1 Gbps (109 bits per second). With a packet size, L, of 1,000 bytes
VideoNote
Developing a protocol
and FSM representation
for a simple applicationlayer protocol
216
CHAPTER 3
•
TRANSPORT LAYER
Sender
send pkt0
Receiver
pkt0
send pkt0
0
ACK
rcv ACK0
send pkt1
rcv pkt1
send ACK1
pkt0
rcv pkt0
send ACK0
0
ACK
rcv ACK0
send pkt1
timeout
resend pkt1
pkt0
0
ACK
Receiver
rcv pkt0
send ACK0
pkt1
1
ACK
rcv ACK1
send pkt0
Sender
pkt1
X (loss)
pkt1
rcv pkt0
send ACK0
1
rcv pkt1
send ACK1
0
rcv pkt0
send ACK0
ACK
rcv ACK1
send pkt0
pkt0
ACK
b. Lost packet
a. Operation with no loss
Sender
send pkt0
Receiver
pkt0
send pkt0
0
ACK
rcv ACK0
send pkt1
Sender
A
rcv pkt1
send ACK1
(loss) X
timeout
resend pkt1
pkt1
1
ACK
rcv ACK1
send pkt0
pkt0
0
ACK
pkt0
rcv pkt0
send ACK0
pkt1
CK1
Receiver
rcv pkt1
(detect
duplicate)
send ACK1
0
ACK
rcv ACK0
send pkt1
pkt1
timeout
resend pkt1
pkt1
rcv ACK1
send pkt0
pkt0
rcv ACK1
do nothing
CK1
rcv pkt0
send ACK0
rcv pkt1
send ACK1
A
1
ACK
0
ACK
rcv pkt 1
(detect duplicate)
send ACK1
rcv pkt0
send ACK0
rcv pkt0
send ACK0
d. Premature timeout
c. Lost ACK
Figure 3.16
Operation of rdt3.0, the alternating-bit protocol
3.4
•
PRINCIPLES OF RELIABLE DATA TRANSFER
Data packet
Data packets
ACK packets
a. A stop-and-wait protocol in operation
Figure 3.17
b. A pipelined protocol in operation
Stop-and-wait versus pipelined protocol
(8,000 bits) per packet, including both header fields and data, the time needed to
actually transmit the packet into the 1 Gbps link is
dtrans =
8000 bits>packet
L
=
= 8 microseconds
R
109 bits/sec
Figure 3.18(a) shows that with our stop-and-wait protocol, if the sender begins
sending the packet at t = 0, then at t = L/R = 8 microseconds, the last bit enters the
channel at the sender side. The packet then makes its 15-msec cross-country journey, with the last bit of the packet emerging at the receiver at t = RTT/2 + L/R =
15.008 msec. Assuming for simplicity that ACK packets are extremely small (so that
we can ignore their transmission time) and that the receiver can send an ACK as
soon as the last bit of a data packet is received, the ACK emerges back at the sender
at t = RTT + L/R = 30.008 msec. At this point, the sender can now transmit the next
message. Thus, in 30.008 msec, the sender was sending for only 0.008 msec. If we
define the utilization of the sender (or the channel) as the fraction of time the sender
is actually busy sending bits into the channel, the analysis in Figure 3.18(a) shows
that the stop-and-wait protocol has a rather dismal sender utilization, Usender, of
Usender =
L>R
RTT + L>R
=
.008
= 0.00027
30.008
That is, the sender was busy only 2.7 hundredths of one percent of the time!
Viewed another way, the sender was able to send only 1,000 bytes in 30.008 milliseconds, an effective throughput of only 267 kbps—even though a 1 Gbps link was
available! Imagine the unhappy network manager who just paid a fortune for a gigabit capacity link but manages to get a throughput of only 267 kilobits per second!
This is a graphic example of how network protocols can limit the capabilities
217
218
CHAPTER 3
•
TRANSPORT LAYER
provided by the underlying network hardware. Also, we have neglected lower-layer
protocol-processing times at the sender and receiver, as well as the processing and
queuing delays that would occur at any intermediate routers between the sender
and receiver. Including these effects would serve only to further increase the delay
and further accentuate the poor performance.
The solution to this particular performance problem is simple: Rather than operate in a stop-and-wait manner, the sender is allowed to send multiple packets without waiting for acknowledgments, as illustrated in Figure 3.17(b). Figure 3.18(b)
shows that if the sender is allowed to transmit three packets before having to wait
for acknowledgments, the utilization of the sender is essentially tripled. Since the
many in-transit sender-to-receiver packets can be visualized as filling a pipeline, this
technique is known as pipelining. Pipelining has the following consequences for
reliable data transfer protocols:
• The range of sequence numbers must be increased, since each in-transit packet
(not counting retransmissions) must have a unique sequence number and there
may be multiple, in-transit, unacknowledged packets.
• The sender and receiver sides of the protocols may have to buffer more than one
packet. Minimally, the sender will have to buffer packets that have been transmitted but not yet acknowledged. Buffering of correctly received packets may
also be needed at the receiver, as discussed below.
• The range of sequence numbers needed and the buffering requirements will
depend on the manner in which a data transfer protocol responds to lost, corrupted, and overly delayed packets. Two basic approaches toward pipelined error
recovery can be identified: Go-Back-N and selective repeat.
3.4.3 Go-Back-N (GBN)
In a Go-Back-N (GBN) protocol, the sender is allowed to transmit multiple packets
(when available) without waiting for an acknowledgment, but is constrained to have no
more than some maximum allowable number, N, of unacknowledged packets in the
pipeline. We describe the GBN protocol in some detail in this section. But before reading on, you are encouraged to play with the GBN applet (an awesome applet!) at the
companion Web site.
Figure 3.19 shows the sender’s view of the range of sequence numbers in a GBN
protocol. If we define base to be the sequence number of the oldest unacknowledged
packet and nextseqnum to be the smallest unused sequence number (that is, the
sequence number of the next packet to be sent), then four intervals in the range of
sequence numbers can be identified. Sequence numbers in the interval [0,base-1]
correspond to packets that have already been transmitted and acknowledged. The interval [base,nextseqnum-1] corresponds to packets that have been sent but not yet
acknowledged. Sequence numbers in the interval [nextseqnum,base+N-1] can
3.4
•
PRINCIPLES OF RELIABLE DATA TRANSFER
Sender
219
Receiver
First bit of first packet
transmitted, t = 0
Last bit of first packet
transmitted, t = L/R
First bit of first packet arrives
RTT
Last bit of first packet arrives, send ACK
ACK arrives, send next packet,
t = RTT + L/R
a. Stop-and-wait operation
Sender
Receiver
First bit of first packet
transmitted, t = 0
Last bit of first packet
transmitted, t = L/R
First bit of first packet arrives
RTT
Last bit of first packet arrives, send ACK
Last bit of 2nd packet arrives, send ACK
Last bit of 3rd packet arrives, send ACK
ACK arrives, send next packet,
t = RTT + L/R
b. Pipelined operation
Figure 3.18
Stop-and-wait and pipelined sending
220
CHAPTER 3
base
•
TRANSPORT LAYER
nextseqnum
Window size
N
Figure 3.19
Key:
Already
ACK’d
Usable,
not yet sent
Sent, not
yet ACK’d
Not usable
Sender’s view of sequence numbers in Go-Back-N
be used for packets that can be sent immediately, should data arrive from the upper
layer. Finally, sequence numbers greater than or equal to base+N cannot be used until
an unacknowledged packet currently in the pipeline (specifically, the packet with
sequence number base) has been acknowledged.
As suggested by Figure 3.19, the range of permissible sequence numbers for
transmitted but not yet acknowledged packets can be viewed as a window of size N
over the range of sequence numbers. As the protocol operates, this window slides
forward over the sequence number space. For this reason, N is often referred to as
the window size and the GBN protocol itself as a sliding-window protocol. You
might be wondering why we would even limit the number of outstanding, unacknowledged packets to a value of N in the first place. Why not allow an unlimited
number of such packets? We’ll see in Section 3.5 that flow control is one reason to
impose a limit on the sender. We’ll examine another reason to do so in Section 3.7,
when we study TCP congestion control.
In practice, a packet’s sequence number is carried in a fixed-length field in the
packet header. If k is the number of bits in the packet sequence number field, the range
of sequence numbers is thus [0,2k – 1]. With a finite range of sequence numbers, all
arithmetic involving sequence numbers must then be done using modulo 2k arithmetic.
(That is, the sequence number space can be thought of as a ring of size 2k, where
sequence number 2k– 1 is immediately followed by sequence number 0.) Recall that
rdt3.0 had a 1-bit sequence number and a range of sequence numbers of [0,1]. Several of the problems at the end of this chapter explore the consequences of a finite range
of sequence numbers. We will see in Section 3.5 that TCP has a 32-bit sequence number
field, where TCP sequence numbers count bytes in the byte stream rather than packets.
Figures 3.20 and 3.21 give an extended FSM description of the sender and
receiver sides of an ACK-based, NAK-free, GBN protocol. We refer to this FSM
description as an extended FSM because we have added variables (similar to programming-language variables) for base and nextseqnum, and added operations
on these variables and conditional actions involving these variables. Note that the
extended FSM specification is now beginning to look somewhat like a programminglanguage specification. [Bochman 1984] provides an excellent survey of additional
extensions to FSM techniques as well as other programming-language-based techniques for specifying protocols.
3.4
PRINCIPLES OF RELIABLE DATA TRANSFER
•
rdt_send(data)
if(nextseqnum<base+N){
sndpkt[nextseqnum]=make_pkt(nextseqnum,data,checksum)
udt_send(sndpkt[nextseqnum])
if(base==nextseqnum)
start_timer
nextseqnum++
}
else
refuse_data(data)
Λ
base=1
nextseqnum=1
timeout
Wait
rdt_rcv(rcvpkt) && corrupt(rcvpkt)
Λ
start_timer
udt_send(sndpkt[base])
udt_send(sndpkt[base+1])
...
udt_send(sndpkt[nextseqnum-1])
rdt_rcv(rcvpkt) && notcorrupt(rcvpkt)
base=getacknum(rcvpkt)+1
If(base==nextseqnum)
stop_timer
else
start_timer
Figure 3.20
Extended FSM description of GBN sender
rdt_rcv(rcvpkt)
&& notcorrupt(rcvpkt)
&& hasseqnum(rcvpkt,expectedseqnum)
extract(rcvpkt,data)
deliver_data(data)
sndpkt=make_pkt(expectedseqnum,ACK,checksum)
udt_send(sndpkt)
expectedseqnum++
Wait
Λ
default
udt_send(sndpkt)
expectedseqnum=1
sndpkt=make_pkt(0,ACK,checksum)
Figure 3.21
Extended FSM description of GBN receiver
221
222
CHAPTER 3
•
TRANSPORT LAYER
The GBN sender must respond to three types of events:
• Invocation from above. When rdt_send() is called from above, the sender
first checks to see if the window is full, that is, whether there are N outstanding,
unacknowledged packets. If the window is not full, a packet is created and sent,
and variables are appropriately updated. If the window is full, the sender simply
returns the data back to the upper layer, an implicit indication that the window is
full. The upper layer would presumably then have to try again later. In a real
implementation, the sender would more likely have either buffered (but not
immediately sent) this data, or would have a synchronization mechanism (for
example, a semaphore or a flag) that would allow the upper layer to call
rdt_send() only when the window is not full.
• Receipt of an ACK. In our GBN protocol, an acknowledgment for a packet with
sequence number n will be taken to be a cumulative acknowledgment, indicating that all packets with a sequence number up to and including n have been correctly received at the receiver. We’ll come back to this issue shortly when we
examine the receiver side of GBN.
• A timeout event. The protocol’s name, “Go-Back-N,” is derived from the sender’s
behavior in the presence of lost or overly delayed packets. As in the stop-and-wait
protocol, a timer will again be used to recover from lost data or acknowledgment
packets. If a timeout occurs, the sender resends all packets that have been previously sent but that have not yet been acknowledged. Our sender in Figure 3.20 uses
only a single timer, which can be thought of as a timer for the oldest transmitted but
not yet acknowledged packet. If an ACK is received but there are still additional
transmitted but not yet acknowledged packets, the timer is restarted. If there are no
outstanding, unacknowledged packets, the timer is stopped.
The receiver’s actions in GBN are also simple. If a packet with sequence number n is received correctly and is in order (that is, the data last delivered to the upper
layer came from a packet with sequence number n – 1), the receiver sends an ACK
for packet n and delivers the data portion of the packet to the upper layer. In all other
cases, the receiver discards the packet and resends an ACK for the most recently
received in-order packet. Note that since packets are delivered one at a time to the
upper layer, if packet k has been received and delivered, then all packets with a
sequence number lower than k have also been delivered. Thus, the use of cumulative acknowledgments is a natural choice for GBN.
In our GBN protocol, the receiver discards out-of-order packets. Although it
may seem silly and wasteful to discard a correctly received (but out-of-order)
packet, there is some justification for doing so. Recall that the receiver must deliver
data in order to the upper layer. Suppose now that packet n is expected, but packet
n + 1 arrives. Because data must be delivered in order, the receiver could buffer
(save) packet n + 1 and then deliver this packet to the upper layer after it had later
3.4
•
PRINCIPLES OF RELIABLE DATA TRANSFER
received and delivered packet n. However, if packet n is lost, both it and packet
n + 1 will eventually be retransmitted as a result of the GBN retransmission rule at
the sender. Thus, the receiver can simply discard packet n + 1. The advantage of this
approach is the simplicity of receiver buffering—the receiver need not buffer any
out-of-order packets. Thus, while the sender must maintain the upper and lower
bounds of its window and the position of nextseqnum within this window, the
only piece of information the receiver need maintain is the sequence number of the
next in-order packet. This value is held in the variable expectedseqnum, shown
in the receiver FSM in Figure 3.21. Of course, the disadvantage of throwing away a
correctly received packet is that the subsequent retransmission of that packet might
be lost or garbled and thus even more retransmissions would be required.
Figure 3.22 shows the operation of the GBN protocol for the case of a window
size of four packets. Because of this window size limitation, the sender sends packets 0 through 3 but then must wait for one or more of these packets to be acknowledged before proceeding. As each successive ACK (for example, ACK0 and ACK1)
is received, the window slides forward and the sender can transmit one new packet
(pkt4 and pkt5, respectively). On the receiver side, packet 2 is lost and thus packets
3, 4, and 5 are found to be out of order and are discarded.
Before closing our discussion of GBN, it is worth noting that an implementation of this protocol in a protocol stack would likely have a structure similar to that
of the extended FSM in Figure 3.20. The implementation would also likely be in the
form of various procedures that implement the actions to be taken in response to the
various events that can occur. In such event-based programming, the various procedures are called (invoked) either by other procedures in the protocol stack, or as
the result of an interrupt. In the sender, these events would be (1) a call from the
upper-layer entity to invoke rdt_send(), (2) a timer interrupt, and (3) a call from
the lower layer to invoke rdt_rcv() when a packet arrives. The programming
exercises at the end of this chapter will give you a chance to actually implement
these routines in a simulated, but realistic, network setting.
We note here that the GBN protocol incorporates almost all of the techniques
that we will encounter when we study the reliable data transfer components of TCP
in Section 3.5. These techniques include the use of sequence numbers, cumulative
acknowledgments, checksums, and a timeout/retransmit operation.
3.4.4 Selective Repeat (SR)
The GBN protocol allows the sender to potentially “fill the pipeline” in Figure 3.17
with packets, thus avoiding the channel utilization problems we noted with stopand-wait protocols. There are, however, scenarios in which GBN itself suffers from
performance problems. In particular, when the window size and bandwidth-delay
product are both large, many packets can be in the pipeline. A single packet error
can thus cause GBN to retransmit a large number of packets, many unnecessarily.
As the probability of channel errors increases, the pipeline can become filled with
223
224
CHAPTER 3
•
TRANSPORT LAYER
Sender
Receiver
send pkt0
send pkt1
rcv pkt0
send ACK0
send pkt2
rcv pkt1
send ACK1
X
(loss)
send pkt3
(wait)
rcv pkt3, discard
send ACK1
rcv
send
rcv
send
ACK0
pkt4
ACK1
pkt5
rcv pkt4, discard
send ACK1
pkt2 timeout
send pkt2
send pkt3
send pkt4
send pkt5
Figure 3.22
rcv pkt5, discard
send ACK1
rcv pkt2, deliver
send ACK2
rcv pkt3, deliver
send ACK3
Go-Back-N in operation
these unnecessary retransmissions. Imagine, in our message-dictation scenario, that
if every time a word was garbled, the surrounding 1,000 words (for example, a window size of 1,000 words) had to be repeated. The dictation would be slowed by all
of the reiterated words.
As the name suggests, selective-repeat protocols avoid unnecessary retransmissions by having the sender retransmit only those packets that it suspects were
received in error (that is, were lost or corrupted) at the receiver. This individual, asneeded, retransmission will require that the receiver individually acknowledge correctly received packets. A window size of N will again be used to limit the number
3.4
send_base
•
PRINCIPLES OF RELIABLE DATA TRANSFER
nextseqnum
Window size
N
Key:
Already
ACK’d
Usable,
not yet sent
Sent, not
yet ACK’d
Not usable
a. Sender view of sequence numbers
rcv_base
Key:
Window size
N
Out of order
(buffered) but
already ACK’d
Acceptable
(within
window)
Expected, not
yet received
Not usable
b. Receiver view of sequence numbers
Figure 3.23
Selective-repeat (SR) sender and receiver views of
sequence-number space
of outstanding, unacknowledged packets in the pipeline. However, unlike GBN, the
sender will have already received ACKs for some of the packets in the window.
Figure 3.23 shows the SR sender’s view of the sequence number space. Figure 3.24
details the various actions taken by the SR sender.
The SR receiver will acknowledge a correctly received packet whether or not it
is in order. Out-of-order packets are buffered until any missing packets (that is,
packets with lower sequence numbers) are received, at which point a batch of packets can be delivered in order to the upper layer. Figure 3.25 itemizes the various
actions taken by the SR receiver. Figure 3.26 shows an example of SR operation in
the presence of lost packets. Note that in Figure 3.26, the receiver initially buffers
packets 3, 4, and 5, and delivers them together with packet 2 to the upper layer when
packet 2 is finally received.
It is important to note that in Step 2 in Figure 3.25, the receiver reacknowledges
(rather than ignores) already received packets with certain sequence numbers below
the current window base. You should convince yourself that this reacknowledgment
is indeed needed. Given the sender and receiver sequence number spaces in Figure
3.23, for example, if there is no ACK for packet send_base propagating from the
receiver to the sender, the sender will eventually retransmit packet send_base,
even though it is clear (to us, not the sender!) that the receiver has already received
225
226
CHAPTER 3
•
TRANSPORT LAYER
1. Data received from above. When data is received from above, the SR sender
checks the next available sequence number for the packet. If the sequence
number is within the sender’s window, the data is packetized and sent; otherwise it is either buffered or returned to the upper layer for later transmission,
as in GBN.
2. Timeout. Timers are again used to protect against lost packets. However, each
packet must now have its own logical timer, since only a single packet will
be transmitted on timeout. A single hardware timer can be used to mimic the
operation of multiple logical timers [Varghese 1997].
3. ACK received. If an ACK is received, the SR sender marks that packet as
having been received, provided it is in the window. If the packet’s sequence
number is equal to send_base, the window base is moved forward to the
unacknowledged packet with the smallest sequence number. If the window
moves and there are untransmitted packets with sequence numbers that now
fall within the window, these packets are transmitted.
Figure 3.24
SR sender events and actions
1. Packet with sequence number in [rcv_base, rcv_base+N-1] is correctly received. In this case, the received packet falls within the receiver’s window and a selective ACK packet is returned to the sender. If the packet was not
previously received, it is buffered. If this packet has a sequence number equal to
the base of the receive window (rcv_base in Figure 3.22), then this packet,
and any previously buffered and consecutively numbered (beginning with
rcv_base) packets are delivered to the upper layer. The receive window is
then moved forward by the number of packets delivered to the upper layer. As
an example, consider Figure 3.26. When a packet with a sequence number of
rcv_base=2 is received, it and packets 3, 4, and 5 can be delivered to the
upper layer.
2. Packet with sequence number in [rcv_base-N, rcv_base-1] is correctly received. In this case, an ACK must be generated, even though this is a
packet that the receiver has previously acknowledged.
3. Otherwise. Ignore the packet.
Figure 3.25
SR receiver events and actions
that packet. If the receiver were not to acknowledge this packet, the sender’s window would never move forward! This example illustrates an important aspect of SR
protocols (and many other protocols as well). The sender and receiver will not
always have an identical view of what has been received correctly and what has not.
For SR protocols, this means that the sender and receiver windows will not always
coincide.
3.4
Sender
•
PRINCIPLES OF RELIABLE DATA TRANSFER
227
Receiver
pkt0 sent
0 1 2 3 4 5 6 7 8 9
pkt0 rcvd, delivered, ACK0 sent
0 1 2 3 4 5 6 7 8 9
pkt1 sent
0 1 2 3 4 5 6 7 8 9
pkt2 sent
0 1 2 3 4 5 6 7 8 9
X
pkt1 rcvd, delivered, ACK1 sent
0 1 2 3 4 5 6 7 8 9
(loss)
pkt3 sent, window full
0 1 2 3 4 5 6 7 8 9
ACK0 rcvd, pkt4 sent
0 1 2 3 4 5 6 7 8 9
ACK1 rcvd, pkt5 sent
0 1 2 3 4 5 6 7 8 9
pkt2 TIMEOUT, pkt2
resent
0 1 2 3 4 5 6 7 8 9
pkt3 rcvd, buffered, ACK3 sent
0 1 2 3 4 5 6 7 8 9
pkt4 rcvd, buffered, ACK4 sent
0 1 2 3 4 5 6 7 8 9
pkt5 rcvd; buffered, ACK5 sent
0 1 2 3 4 5 6 7 8 9
pkt2 rcvd, pkt2,pkt3,pkt4,pkt5
delivered, ACK2 sent
0 1 2 3 4 5 6 7 8 9
ACK3 rcvd, nothing sent
0 1 2 3 4 5 6 7 8 9
Figure 3.26
SR operation
The lack of synchronization between sender and receiver windows has important consequences when we are faced with the reality of a finite range of sequence
numbers. Consider what could happen, for example, with a finite range of four packet
sequence numbers, 0, 1, 2, 3, and a window size of three. Suppose packets 0 through
2 are transmitted and correctly received and acknowledged at the receiver. At this
point, the receiver’s window is over the fourth, fifth, and sixth packets, which have
sequence numbers 3, 0, and 1, respectively. Now consider two scenarios. In the first
scenario, shown in Figure 3.27(a), the ACKs for the first three packets are lost and
228
CHAPTER 3
•
TRANSPORT LAYER
Sender window
(after receipt)
Receiver window
(after receipt)
0 1 2 3 0 1 2
pkt0
0 1 2 3 0 1 2
pkt1
0 1 2 3 0 1 2
pkt2
x
ACK0
0 1 2 3 0 1 2
ACK1
0 1 2 3 0 1 2
ACK2
0 1 2 3 0 1 2
x
timeout
retransmit pkt0
0 1 2 3 0 1 2
x
pkt0
receive packet
with seq number 0
a.
Sender window
(after receipt)
Receiver window
(after receipt)
0 1 2 3 0 1 2
pkt0
0 1 2 3 0 1 2
pkt1
0 1 2 3 0 1 2
pkt2
0 1 2 3 0 1 2
pkt3
0 1 2 3 0 1 2
pkt0
ACK0
0 1 2 3 0 1 2
ACK1
0 1 2 3 0 1 2
ACK2
0 1 2 3 0 1 2
x
receive packet
with seq number 0
b.
Figure 3.27
SR receiver dilemma with too-large windows: A new packet
or a retransmission?
3.4
•
PRINCIPLES OF RELIABLE DATA TRANSFER
the sender retransmits these packets. The receiver thus next receives a packet with
sequence number 0—a copy of the first packet sent.
In the second scenario, shown in Figure 3.27(b), the ACKs for the first three
packets are all delivered correctly. The sender thus moves its window forward and
sends the fourth, fifth, and sixth packets, with sequence numbers 3, 0, and 1, respectively. The packet with sequence number 3 is lost, but the packet with sequence
number 0 arrives—a packet containing new data.
Now consider the receiver’s viewpoint in Figure 3.27, which has a figurative
curtain between the sender and the receiver, since the receiver cannot “see” the
actions taken by the sender. All the receiver observes is the sequence of messages it
receives from the channel and sends into the channel. As far as it is concerned, the
two scenarios in Figure 3.27 are identical. There is no way of distinguishing the
retransmission of the first packet from an original transmission of the fifth packet.
Clearly, a window size that is 1 less than the size of the sequence number space
won’t work. But how small must the window size be? A problem at the end of the
chapter asks you to show that the window size must be less than or equal to half the
size of the sequence number space for SR protocols.
At the companion Web site, you will find an applet that animates the operation
of the SR protocol. Try performing the same experiments that you did with the GBN
applet. Do the results agree with what you expect?
This completes our discussion of reliable data transfer protocols. We’ve covered
a lot of ground and introduced numerous mechanisms that together provide for reliable data transfer. Table 3.1 summarizes these mechanisms. Now that we have seen all
of these mechanisms in operation and can see the “big picture,” we encourage you to
review this section again to see how these mechanisms were incrementally added to
cover increasingly complex (and realistic) models of the channel connecting the
sender and receiver, or to improve the performance of the protocols.
Let’s conclude our discussion of reliable data transfer protocols by considering
one remaining assumption in our underlying channel model. Recall that we have
assumed that packets cannot be reordered within the channel between the sender and
receiver. This is generally a reasonable assumption when the sender and receiver are
connected by a single physical wire. However, when the “channel” connecting the two
is a network, packet reordering can occur. One manifestation of packet reordering is
that old copies of a packet with a sequence or acknowledgment number of x can
appear, even though neither the sender’s nor the receiver’s window contains x. With
packet reordering, the channel can be thought of as essentially buffering packets and
spontaneously emitting these packets at any point in the future. Because sequence
numbers may be reused, some care must be taken to guard against such duplicate
packets. The approach taken in practice is to ensure that a sequence number is not
reused until the sender is “sure” that any previously sent packets with sequence number x are no longer in the network. This is done by assuming that a packet cannot
“live” in the network for longer than some fixed maximum amount of time. A maximum packet lifetime of approximately three minutes is assumed in the TCP extensions
229
230
CHAPTER 3
•
TRANSPORT LAYER
Mechanism
Use, Comments
Checksum
Used to detect bit errors in a transmitted packet.
Timer
Used to timeout/retransmit a packet, possibly because the packet (or its ACK) was
lost within the channel. Because timeouts can occur when a packet is delayed but
not lost (premature timeout), or when a packet has been received by the receiver
but the receiver-to-sender ACK has been lost, duplicate copies of a packet may be
received by a receiver.
Sequence number
Used for sequential numbering of packets of data flowing from sender to receiver.
Gaps in the sequence numbers of received packets allow the receiver to detect a
lost packet. Packets with duplicate sequence numbers allow the receiver to detect
duplicate copies of a packet.
Acknowledgment
Used by the receiver to tell the sender that a packet or set of packets has been
received correctly. Acknowledgments will typically carry the sequence number of the
packet or packets being acknowledged. Acknowledgments may be individual or
cumulative, depending on the protocol.
Negative acknowledgment
Used by the receiver to tell the sender that a packet has not been received correctly. Negative acknowledgments will typically carry the sequence number of the packet that was not received correctly.
Window, pipelining
The sender may be restricted to sending only packets with sequence numbers that
fall within a given range. By allowing multiple packets to be transmitted but not yet
acknowledged, sender utilization can be increased over a stop-and-wait mode of
operation. We’ll see shortly that the window size may be set on the basis of the
receiver’s ability to receive and buffer messages, or the level of congestion in the
network, or both.
Table 3.1
Summary of reliable data transfer mechanisms and their use
for high-speed networks [RFC 1323]. [Sunshine 1978] describes a method for using
sequence numbers such that reordering problems can be completely avoided.
3.5 Connection-Oriented Transport: TCP
Now that we have covered the underlying principles of reliable data transfer, let’s
turn to TCP—the Internet’s transport-layer, connection-oriented, reliable transport
protocol. In this section, we’ll see that in order to provide reliable data transfer, TCP
relies on many of the underlying principles discussed in the previous section,
including error detection, retransmissions, cumulative acknowledgments, timers,
3.5
•
CONNECTION-ORIENTED TRANSPORT: TCP
and header fields for sequence and acknowledgment numbers. TCP is defined in
RFC 793, RFC 1122, RFC 1323, RFC 2018, and RFC 2581.
3.5.1 The TCP Connection
TCP is said to be connection-oriented because before one application process can
begin to send data to another, the two processes must first “handshake” with each
other—that is, they must send some preliminary segments to each other to establish the
parameters of the ensuing data transfer. As part of TCP connection establishment, both
sides of the connection will initialize many TCP state variables (many of which will be
discussed in this section and in Section 3.7) associated with the TCP connection.
The TCP “connection” is not an end-to-end TDM or FDM circuit as in a circuitswitched network. Nor is it a virtual circuit (see Chapter 1), as the connection state
resides entirely in the two end systems. Because the TCP protocol runs only in the
end systems and not in the intermediate network elements (routers and link-layer
switches), the intermediate network elements do not maintain TCP connection state.
CASE HISTORY
VINTON CERF, ROBERT KAHN, AND TCP/IP
In the early 1970s, packet-switched networks began to proliferate, with the
ARPAnet—the precursor of the Internet—being just one of many networks. Each of
these networks had its own protocol. Two researchers, Vinton Cerf and Robert Kahn,
recognized the importance of interconnecting these networks and invented a crossnetwork protocol called TCP/IP, which stands for Transmission Control
Protocol/Internet Protocol. Although Cerf and Kahn began by seeing the protocol as
a single entity, it was later split into its two parts, TCP and IP, which operated separately. Cerf and Kahn published a paper on TCP/IP in May 1974 in IEEE
Transactions on Communications Technology [Cerf 1974].
The TCP/IP protocol, which is the bread and butter of today’s Internet, was devised
before PCs, workstations, smartphones, and tablets, before the proliferation of Ethernet,
cable, and DSL, WiFi, and other access network technologies, and before the Web,
social media, and streaming video. Cerf and Kahn saw the need for a networking protocol that, on the one hand, provides broad support for yet-to-be-defined applications
and, on the other hand, allows arbitrary hosts and link-layer protocols to interoperate.
In 2004, Cerf and Kahn received the ACM’s Turing Award, considered the
“Nobel Prize of Computing” for “pioneering work on internetworking, including the
design and implementation of the Internet’s basic communications protocols, TCP/IP,
and for inspired leadership in networking.”
231
232
CHAPTER 3
•
TRANSPORT LAYER
In fact, the intermediate routers are completely oblivious to TCP connections; they
see datagrams, not connections.
A TCP connection provides a full-duplex service: If there is a TCP connection
between Process A on one host and Process B on another host, then applicationlayer data can flow from Process A to Process B at the same time as applicationlayer data flows from Process B to Process A. A TCP connection is also always
point-to-point, that is, between a single sender and a single receiver. So-called
“multicasting” (see Section 4.7)—the transfer of data from one sender to many
receivers in a single send operation—is not possible with TCP. With TCP, two hosts
are company and three are a crowd!
Let’s now take a look at how a TCP connection is established. Suppose a
process running in one host wants to initiate a connection with another process in
another host. Recall that the process that is initiating the connection is called the
client process, while the other process is called the server process. The client application process first informs the client transport layer that it wants to establish a
connection to a process in the server. Recall from Section 2.7.2, a Python client program does this by issuing the command
clientSocket.connect((serverName,serverPort))
where serverName is the name of the server and serverPort identifies the
process on the server. TCP in the client then proceeds to establish a TCP connection
with TCP in the server. At the end of this section we discuss in some detail the connection-establishment procedure. For now it suffices to know that the client first sends
a special TCP segment; the server responds with a second special TCP segment; and
finally the client responds again with a third special segment. The first two segments
carry no payload, that is, no application-layer data; the third of these segments may
carry a payload. Because three segments are sent between the two hosts, this connection-establishment procedure is often referred to as a three-way handshake.
Once a TCP connection is established, the two application processes can send
data to each other. Let’s consider the sending of data from the client process to the
server process. The client process passes a stream of data through the socket (the
door of the process), as described in Section 2.7. Once the data passes through
the door, the data is in the hands of TCP running in the client. As shown in Figure
3.28, TCP directs this data to the connection’s send buffer, which is one of the
buffers that is set aside during the initial three-way handshake. From time to time,
TCP will grab chunks of data from the send buffer and pass the data to the network
layer. Interestingly, the TCP specification [RFC 793] is very laid back about specifying when TCP should actually send buffered data, stating that TCP should “send
that data in segments at its own convenience.” The maximum amount of data that
can be grabbed and placed in a segment is limited by the maximum segment size
(MSS). The MSS is typically set by first determining the length of the largest
link-layer frame that can be sent by the local sending host (the so-called maximum
3.5
•
CONNECTION-ORIENTED TRANSPORT: TCP
Process
writes data
Process
reads data
Socket
Socket
Segment
TCP
send
buffer
Figure 3.28
Segment
TCP
receive
buffer
TCP send and receive buffers
transmission unit, MTU), and then setting the MSS to ensure that a TCP segment
(when encapsulated in an IP datagram) plus the TCP/IP header length (typically 40
bytes) will fit into a single link-layer frame. Both Ethernet and PPP link-layer protocols have an MSS of 1,500 bytes. Approaches have also been proposed for discovering the path MTU—the largest link-layer frame that can be sent on all links from
source to destination [RFC 1191]—and setting the MSS based on the path MTU
value. Note that the MSS is the maximum amount of application-layer data in the
segment, not the maximum size of the TCP segment including headers. (This terminology is confusing, but we have to live with it, as it is well entrenched.)
TCP pairs each chunk of client data with a TCP header, thereby forming TCP
segments. The segments are passed down to the network layer, where they are separately encapsulated within network-layer IP datagrams. The IP datagrams are then
sent into the network. When TCP receives a segment at the other end, the segment’s
data is placed in the TCP connection’s receive buffer, as shown in Figure 3.28. The
application reads the stream of data from this buffer. Each side of the connection has
its own send buffer and its own receive buffer. (You can see the online flow-control
applet at http://www.awl.com/kurose-ross, which provides an animation of the send
and receive buffers.)
We see from this discussion that a TCP connection consists of buffers, variables, and a socket connection to a process in one host, and another set of buffers,
variables, and a socket connection to a process in another host. As mentioned earlier, no buffers or variables are allocated to the connection in the network elements
(routers, switches, and repeaters) between the hosts.
3.5.2 TCP Segment Structure
Having taken a brief look at the TCP connection, let’s examine the TCP segment
structure. The TCP segment consists of header fields and a data field. The data
field contains a chunk of application data. As mentioned above, the MSS limits the
233
CHAPTER 3
•
TRANSPORT LAYER
32 bits
Source port #
Dest port #
Sequence number
Acknowledgment number
Header
Unused
length
URG
ACK
PSH
RST
SYN
FIN
234
Internet checksum
Receive window
Urgent data pointer
Options
Data
Figure 3.29
TCP segment structure
maximum size of a segment’s data field. When TCP sends a large file, such as an
image as part of a Web page, it typically breaks the file into chunks of size MSS
(except for the last chunk, which will often be less than the MSS). Interactive applications, however, often transmit data chunks that are smaller than the MSS; for
example, with remote login applications like Telnet, the data field in the TCP segment is often only one byte. Because the TCP header is typically 20 bytes (12 bytes
more than the UDP header), segments sent by Telnet may be only 21 bytes in length.
Figure 3.29 shows the structure of the TCP segment. As with UDP, the header
includes source and destination port numbers, which are used for
multiplexing/demultiplexing data from/to upper-layer applications. Also, as with
UDP, the header includes a checksum field. A TCP segment header also contains
the following fields:
• The 32-bit sequence number field and the 32-bit acknowledgment number
field are used by the TCP sender and receiver in implementing a reliable data
transfer service, as discussed below.
• The 16-bit receive window field is used for flow control. We will see shortly that
it is used to indicate the number of bytes that a receiver is willing to accept.
• The 4-bit header length field specifies the length of the TCP header in 32-bit
words. The TCP header can be of variable length due to the TCP options field.
3.5
•
CONNECTION-ORIENTED TRANSPORT: TCP
(Typically, the options field is empty, so that the length of the typical TCP header
is 20 bytes.)
• The optional and variable-length options field is used when a sender and
receiver negotiate the maximum segment size (MSS) or as a window scaling factor for use in high-speed networks. A time-stamping option is also defined. See
RFC 854 and RFC 1323 for additional details.
• The flag field contains 6 bits. The ACK bit is used to indicate that the value carried in the acknowledgment field is valid; that is, the segment contains an
acknowledgment for a segment that has been successfully received. The RST,
SYN, and FIN bits are used for connection setup and teardown, as we will discuss at the end of this section. Setting the PSH bit indicates that the receiver
should pass the data to the upper layer immediately. Finally, the URG bit is used
to indicate that there is data in this segment that the sending-side upper-layer
entity has marked as “urgent.” The location of the last byte of this urgent data is
indicated by the 16-bit urgent data pointer field. TCP must inform the receiving-side upper-layer entity when urgent data exists and pass it a pointer to the
end of the urgent data. (In practice, the PSH, URG, and the urgent data pointer
are not used. However, we mention these fields for completeness.)
Sequence Numbers and Acknowledgment Numbers
Two of the most important fields in the TCP segment header are the sequence number
field and the acknowledgment number field. These fields are a critical part of TCP’s
reliable data transfer service. But before discussing how these fields are used to provide
reliable data transfer, let us first explain what exactly TCP puts in these fields.
TCP views data as an unstructured, but ordered, stream of bytes. TCP’s use of
sequence numbers reflects this view in that sequence numbers are over the stream of
transmitted bytes and not over the series of transmitted segments. The sequence
number for a segment is therefore the byte-stream number of the first byte in the
segment. Let’s look at an example. Suppose that a process in Host A wants to send a
stream of data to a process in Host B over a TCP connection. The TCP in Host A will
implicitly number each byte in the data stream. Suppose that the data stream consists
of a file consisting of 500,000 bytes, that the MSS is 1,000 bytes, and that the first
byte of the data stream is numbered 0. As shown in Figure 3.30, TCP constructs 500
segments out of the data stream. The first segment gets assigned sequence number 0,
the second segment gets assigned sequence number 1,000, the third segment gets
assigned sequence number 2,000, and so on. Each sequence number is inserted in the
sequence number field in the header of the appropriate TCP segment.
Now let’s consider acknowledgment numbers. These are a little trickier than
sequence numbers. Recall that TCP is full-duplex, so that Host A may be receiving
data from Host B while it sends data to Host B (as part of the same TCP connection).
Each of the segments that arrive from Host B has a sequence number for the data
235
236
CHAPTER 3
•
TRANSPORT LAYER
File
Data for 1st segment
0
Figure 3.30
1
Data for 2nd segment
1,000
1,999
499,999
Dividing file data into TCP segments
flowing from B to A. The acknowledgment number that Host A puts in its segment
is the sequence number of the next byte Host A is expecting from Host B. It is good
to look at a few examples to understand what is going on here. Suppose that Host A
has received all bytes numbered 0 through 535 from B and suppose that it is about
to send a segment to Host B. Host A is waiting for byte 536 and all the subsequent
bytes in Host B’s data stream. So Host A puts 536 in the acknowledgment number
field of the segment it sends to B.
As another example, suppose that Host A has received one segment from Host
B containing bytes 0 through 535 and another segment containing bytes 900 through
1,000. For some reason Host A has not yet received bytes 536 through 899. In this
example, Host A is still waiting for byte 536 (and beyond) in order to re-create B’s
data stream. Thus, A’s next segment to B will contain 536 in the acknowledgment
number field. Because TCP only acknowledges bytes up to the first missing byte in
the stream, TCP is said to provide cumulative acknowledgments.
This last example also brings up an important but subtle issue. Host A received
the third segment (bytes 900 through 1,000) before receiving the second segment
(bytes 536 through 899). Thus, the third segment arrived out of order. The subtle
issue is: What does a host do when it receives out-of-order segments in a TCP connection? Interestingly, the TCP RFCs do not impose any rules here and leave the
decision up to the people programming a TCP implementation. There are basically
two choices: either (1) the receiver immediately discards out-of-order segments
(which, as we discussed earlier, can simplify receiver design), or (2) the receiver
keeps the out-of-order bytes and waits for the missing bytes to fill in the gaps.
Clearly, the latter choice is more efficient in terms of network bandwidth, and is the
approach taken in practice.
In Figure 3.30, we assumed that the initial sequence number was zero. In truth,
both sides of a TCP connection randomly choose an initial sequence number. This is
done to minimize the possibility that a segment that is still present in the network
from an earlier, already-terminated connection between two hosts is mistaken for a
valid segment in a later connection between these same two hosts (which also happen to be using the same port numbers as the old connection) [Sunshine 1978].
3.5
•
CONNECTION-ORIENTED TRANSPORT: TCP
Telnet: A Case Study for Sequence and Acknowledgment Numbers
Telnet, defined in RFC 854, is a popular application-layer protocol used for
remote login. It runs over TCP and is designed to work between any pair of hosts.
Unlike the bulk data transfer applications discussed in Chapter 2, Telnet is an
interactive application. We discuss a Telnet example here, as it nicely illustrates
TCP sequence and acknowledgment numbers. We note that many users now
prefer to use the SSH protocol rather than Telnet, since data sent in a Telnet connection (including passwords!) is not encrypted, making Telnet vulnerable to
eavesdropping attacks (as discussed in Section 8.7).
Suppose Host A initiates a Telnet session with Host B. Because Host A initiates
the session, it is labeled the client, and Host B is labeled the server. Each character
typed by the user (at the client) will be sent to the remote host; the remote host will
send back a copy of each character, which will be displayed on the Telnet user’s
screen. This “echo back” is used to ensure that characters seen by the Telnet user
have already been received and processed at the remote site. Each character thus
traverses the network twice between the time the user hits the key and the time the
character is displayed on the user’s monitor.
Now suppose the user types a single letter, ‘C,’ and then grabs a coffee. Let’s examine the TCP segments that are sent between the client and server. As shown in Figure
3.31, we suppose the starting sequence numbers are 42 and 79 for the client and server,
respectively. Recall that the sequence number of a segment is the sequence number of
the first byte in the data field. Thus, the first segment sent from the client will have
sequence number 42; the first segment sent from the server will have sequence number
79. Recall that the acknowledgment number is the sequence number of the next byte of
data that the host is waiting for. After the TCP connection is established but before any
data is sent, the client is waiting for byte 79 and the server is waiting for byte 42.
As shown in Figure 3.31, three segments are sent. The first segment is sent from
the client to the server, containing the 1-byte ASCII representation of the letter ‘C’
in its data field. This first segment also has 42 in its sequence number field, as we
just described. Also, because the client has not yet received any data from the server,
this first segment will have 79 in its acknowledgment number field.
The second segment is sent from the server to the client. It serves a dual purpose. First it provides an acknowledgment of the data the server has received. By
putting 43 in the acknowledgment field, the server is telling the client that it has successfully received everything up through byte 42 and is now waiting for bytes 43
onward. The second purpose of this segment is to echo back the letter ‘C.’ Thus, the
second segment has the ASCII representation of ‘C’ in its data field. This second
segment has the sequence number 79, the initial sequence number of the server-toclient data flow of this TCP connection, as this is the very first byte of data that the
server is sending. Note that the acknowledgment for client-to-server data is carried
in a segment carrying server-to-client data; this acknowledgment is said to be
piggybacked on the server-to-client data segment.
237
238
CHAPTER 3
•
TRANSPORT LAYER
Host A
Host B
Seq=
User types
'C'
42,
Seq=
Host ACKs
receipt of
echoed 'C'
Time
Figure 3.31
ACK=
79,
Seq=
A
79,
data
3,
CK=4
43,
ACK=
='C'
data
='C'
Host ACKs
receipt of 'C',
echoes back 'C'
80
Time
Sequence and acknowledgment numbers for a simple
Telnet application over TCP
The third segment is sent from the client to the server. Its sole purpose is to
acknowledge the data it has received from the server. (Recall that the second segment contained data—the letter ‘C’—from the server to the client.) This segment
has an empty data field (that is, the acknowledgment is not being piggybacked with
any client-to-server data). The segment has 80 in the acknowledgment number field
because the client has received the stream of bytes up through byte sequence number 79 and it is now waiting for bytes 80 onward. You might think it odd that this
segment also has a sequence number since the segment contains no data. But
because TCP has a sequence number field, the segment needs to have some
sequence number.
3.5.3 Round-Trip Time Estimation and Timeout
TCP, like our rdt protocol in Section 3.4, uses a timeout/retransmit mechanism to
recover from lost segments. Although this is conceptually simple, many subtle
issues arise when we implement a timeout/retransmit mechanism in an actual protocol such as TCP. Perhaps the most obvious question is the length of the timeout
3.5
•
CONNECTION-ORIENTED TRANSPORT: TCP
intervals. Clearly, the timeout should be larger than the connection’s round-trip time
(RTT), that is, the time from when a segment is sent until it is acknowledged. Otherwise, unnecessary retransmissions would be sent. But how much larger? How
should the RTT be estimated in the first place? Should a timer be associated with
each and every unacknowledged segment? So many questions! Our discussion in
this section is based on the TCP work in [Jacobson 1988] and the current IETF recommendations for managing TCP timers [RFC 6298].
Estimating the Round-Trip Time
Let’s begin our study of TCP timer management by considering how TCP estimates
the round-trip time between sender and receiver. This is accomplished as follows.
The sample RTT, denoted SampleRTT, for a segment is the amount of time
between when the segment is sent (that is, passed to IP) and when an acknowledgment for the segment is received. Instead of measuring a SampleRTT for every
transmitted segment, most TCP implementations take only one SampleRTT measurement at a time. That is, at any point in time, the SampleRTT is being estimated
for only one of the transmitted but currently unacknowledged segments, leading to a
new value of SampleRTT approximately once every RTT. Also, TCP never computes a SampleRTT for a segment that has been retransmitted; it only measures
SampleRTT for segments that have been transmitted once [Karn 1987]. (A problem at the end of the chapter asks you to consider why.)
Obviously, the SampleRTT values will fluctuate from segment to segment due
to congestion in the routers and to the varying load on the end systems. Because of
this fluctuation, any given SampleRTT value may be atypical. In order to estimate
a typical RTT, it is therefore natural to take some sort of average of the SampleRTT values. TCP maintains an average, called EstimatedRTT, of the SampleRTT values. Upon obtaining a new SampleRTT, TCP updates
EstimatedRTT according to the following formula:
EstimatedRTT = (1 – ␣) • EstimatedRTT + ␣ • SampleRTT
The formula above is written in the form of a programming-language statement—
the new value of EstimatedRTT is a weighted combination of the previous value
of EstimatedRTT and the new value for SampleRTT. The recommended value
of ␣ is ␣ = 0.125 (that is, 1/8) [RFC 6298], in which case the formula above
becomes:
EstimatedRTT = 0.875 • EstimatedRTT + 0.125 • SampleRTT
Note that EstimatedRTT is a weighted average of the SampleRTT values.
As discussed in a homework problem at the end of this chapter, this weighted average puts more weight on recent samples than on old samples. This is natural, as the
239
240
CHAPTER 3
•
TRANSPORT LAYER
PRINCIPLES IN PRACTICE
TCP provides reliable data transfer by using positive acknowledgments and timers in much
the same way that we studied in Section 3.4. TCP acknowledges data that has been
received correctly, and it then retransmits segments when segments or their corresponding
acknowledgments are thought to be lost or corrupted. Certain versions of TCP also have an
implicit NAK mechanism—with TCP’s fast retransmit mechanism, the receipt of three duplicate ACKs for a given segment serves as an implicit NAK for the following segment, triggering retransmission of that segment before timeout. TCP uses sequences of numbers to
allow the receiver to identify lost or duplicate segments. Just as in the case of our reliable
data transfer protocol, rdt3.0, TCP cannot itself tell for certain if a segment, or its
ACK, is lost, corrupted, or overly delayed. At the sender, TCP’s response will be the same:
retransmit the segment in question.
TCP also uses pipelining, allowing the sender to have multiple transmitted but yet-to-beacknowledged segments outstanding at any given time. We saw earlier that pipelining
can greatly improve a session’s throughput when the ratio of the segment size to roundtrip delay is small. The specific number of outstanding, unacknowledged segments that a
sender can have is determined by TCP’s flow-control and congestion-control mechanisms.
TCP flow control is discussed at the end of this section; TCP congestion control is discussed in Section 3.7. For the time being, we must simply be aware that the TCP sender
uses pipelining.
more recent samples better reflect the current congestion in the network. In statistics, such an average is called an exponential weighted moving average (EWMA).
The word “exponential” appears in EWMA because the weight of a given SampleRTT decays exponentially fast as the updates proceed. In the homework problems you will be asked to derive the exponential term in EstimatedRTT.
Figure 3.32 shows the SampleRTT values and EstimatedRTT for a value of ␣
= 1/8 for a TCP connection between gaia.cs.umass.edu (in Amherst, Massachusetts) to fantasia.eurecom.fr (in the south of France). Clearly, the variations in
the SampleRTT are smoothed out in the computation of the EstimatedRTT.
In addition to having an estimate of the RTT, it is also valuable to have a
measure of the variability of the RTT. [RFC 6298] defines the RTT variation,
DevRTT, as an estimate of how much SampleRTT typically deviates from
EstimatedRTT:
DevRTT = (1 – ␤) • DevRTT + ␤ •| SampleRTT – EstimatedRTT |
Note that DevRTT is an EWMA of the difference between SampleRTT and
EstimatedRTT. If the SampleRTT values have little fluctuation, then DevRTT
will be small; on the other hand, if there is a lot of fluctuation, DevRTT will be
large. The recommended value of β is 0.25.
3.5
•
CONNECTION-ORIENTED TRANSPORT: TCP
350
RTT (milliseconds)
300
Sample RTT
250
200
Estimated RTT
150
100
1
8
15
22
29
36
43
50
57
64
71
78
85
Time (seconds)
Figure 3.32
RTT samples and RTT estimates
Setting and Managing the Retransmission Timeout Interval
Given values of EstimatedRTT and DevRTT, what value should be used for
TCP’s timeout interval? Clearly, the interval should be greater than or equal to
EstimatedRTT, or unnecessary retransmissions would be sent. But the timeout
interval should not be too much larger than EstimatedRTT; otherwise, when a segment is lost, TCP would not quickly retransmit the segment, leading to large data transfer delays. It is therefore desirable to set the timeout equal to the EstimatedRTT plus
some margin. The margin should be large when there is a lot of fluctuation in the
SampleRTT values; it should be small when there is little fluctuation. The value of
DevRTT should thus come into play here. All of these considerations are taken into
account in TCP’s method for determining the retransmission timeout interval:
TimeoutInterval = EstimatedRTT + 4 • DevRTT
An initial TimeoutInterval value of 1 second is recommended [RFC 6298].
Also, when a timeout occurs, the value of TimeoutInterval is doubled to avoid
a premature timeout occurring for a subsequent segment that will soon be acknowledged. However, as soon as a segment is received and EstimatedRTT is updated,
the TimeoutInterval is again computed using the formula above.
92
99
106
241
242
CHAPTER 3
•
TRANSPORT LAYER
3.5.4 Reliable Data Transfer
Recall that the Internet’s network-layer service (IP service) is unreliable. IP does
not guarantee datagram delivery, does not guarantee in-order delivery of datagrams, and does not guarantee the integrity of the data in the datagrams. With IP
service, datagrams can overflow router buffers and never reach their destination,
datagrams can arrive out of order, and bits in the datagram can get corrupted
(flipped from 0 to 1 and vice versa). Because transport-layer segments are carried
across the network by IP datagrams, transport-layer segments can suffer from these
problems as well.
TCP creates a reliable data transfer service on top of IP’s unreliable besteffort service. TCP’s reliable data transfer service ensures that the data stream that a
process reads out of its TCP receive buffer is uncorrupted, without gaps, without
duplication, and in sequence; that is, the byte stream is exactly the same byte stream
that was sent by the end system on the other side of the connection. How TCP provides a reliable data transfer involves many of the principles that we studied in
Section 3.4.
In our earlier development of reliable data transfer techniques, it was conceptually easiest to assume that an individual timer is associated with each transmitted
but not yet acknowledged segment. While this is great in theory, timer management
can require considerable overhead. Thus, the recommended TCP timer management
procedures [RFC 6298] use only a single retransmission timer, even if there are multiple transmitted but not yet acknowledged segments. The TCP protocol described
in this section follows this single-timer recommendation.
We will discuss how TCP provides reliable data transfer in two incremental
steps. We first present a highly simplified description of a TCP sender that uses only
timeouts to recover from lost segments; we then present a more complete description that uses duplicate acknowledgments in addition to timeouts. In the ensuing discussion, we suppose that data is being sent in only one direction, from Host A to
Host B, and that Host A is sending a large file.
Figure 3.33 presents a highly simplified description of a TCP sender. We see
that there are three major events related to data transmission and retransmission in
the TCP sender: data received from application above; timer timeout; and ACK
receipt. Upon the occurrence of the first major event, TCP receives data from the
application, encapsulates the data in a segment, and passes the segment to IP. Note
that each segment includes a sequence number that is the byte-stream number of
the first data byte in the segment, as described in Section 3.5.2. Also note that if the
timer is already not running for some other segment, TCP starts the timer when the
segment is passed to IP. (It is helpful to think of the timer as being associated with
the oldest unacknowledged segment.) The expiration interval for this timer is the
TimeoutInterval, which is calculated from EstimatedRTT and DevRTT,
as described in Section 3.5.3.
3.5
•
CONNECTION-ORIENTED TRANSPORT: TCP
243
/* Assume sender is not constrained by TCP flow or congestion control, that data from above is less
than MSS in size, and that data transfer is in one direction only. */
NextSeqNum=InitialSeqNumber
SendBase=InitialSeqNumber
loop (forever) {
switch(event)
event: data received from application above
create TCP segment with sequence number NextSeqNum
if (timer currently not running)
start timer
pass segment to IP
NextSeqNum=NextSeqNum+length(data)
break;
event: timer timeout
retransmit not-yet-acknowledged segment with
smallest sequence number
start timer
break;
event: ACK received, with ACK field value of y
if (y > SendBase) {
SendBase=y
if (there are currently any not-yet-acknowledged segments)
start timer
}
break;
} /* end of loop forever */
Figure 3.33
Simplified TCP sender
The second major event is the timeout. TCP responds to the timeout event by
retransmitting the segment that caused the timeout. TCP then restarts the timer.
The third major event that must be handled by the TCP sender is the arrival of an
acknowledgment segment (ACK) from the receiver (more specifically, a segment containing a valid ACK field value). On the occurrence of this event, TCP compares the
ACK value y with its variable SendBase. The TCP state variable SendBase is the
sequence number of the oldest unacknowledged byte. (Thus SendBase–1 is the
sequence number of the last byte that is known to have been received correctly and in
order at the receiver.) As indicated earlier, TCP uses cumulative acknowledgments, so
that y acknowledges the receipt of all bytes before byte number y. If y > SendBase,
244
CHAPTER 3
•
TRANSPORT LAYER
then the ACK is acknowledging one or more previously unacknowledged segments.
Thus the sender updates its SendBase variable; it also restarts the timer if there currently are any not-yet-acknowledged segments.
A Few Interesting Scenarios
We have just described a highly simplified version of how TCP provides reliable
data transfer. But even this highly simplified version has many subtleties. To get a
good feeling for how this protocol works, let’s now walk through a few simple
scenarios. Figure 3.34 depicts the first scenario, in which Host A sends one segment to Host B. Suppose that this segment has sequence number 92 and contains 8
bytes of data. After sending this segment, Host A waits for a segment from B with
acknowledgment number 100. Although the segment from A is received at B, the
acknowledgment from B to A gets lost. In this case, the timeout event occurs, and
Host A retransmits the same segment. Of course, when Host B receives the
retransmission, it observes from the sequence number that the segment contains
data that has already been received. Thus, TCP in Host B will discard the bytes in
the retransmitted segment.
Host A
Host B
Seq=9
2, 8
bytes
data
Timeout
ACK=1
00
X
(loss)
Seq=9
2, 8
bytes
ACK=1
Time
Figure 3.34
data
00
Time
Retransmission due to a lost acknowledgment
3.5
•
CONNECTION-ORIENTED TRANSPORT: TCP
In a second scenario, shown in Figure 3.35, Host A sends two segments back to
back. The first segment has sequence number 92 and 8 bytes of data, and the second
segment has sequence number 100 and 20 bytes of data. Suppose that both segments
arrive intact at B, and B sends two separate acknowledgments for each of these segments. The first of these acknowledgments has acknowledgment number 100; the
second has acknowledgment number 120. Suppose now that neither of the acknowledgments arrives at Host A before the timeout. When the timeout event occurs, Host
A resends the first segment with sequence number 92 and restarts the timer. As long
as the ACK for the second segment arrives before the new timeout, the second segment will not be retransmitted.
In a third and final scenario, suppose Host A sends the two segments, exactly as
in the second example. The acknowledgment of the first segment is lost in the
network, but just before the timeout event, Host A receives an acknowledgment with
acknowledgment number 120. Host A therefore knows that Host B has received
everything up through byte 119; so Host A does not resend either of the two
segments. This scenario is illustrated in Figure 3.36.
Host A
Host B
Seq=9
2, 8
bytes
Seq=1
seq=92 timeout interval
data
00, 2
0 byt
es da
ta
A
Seq
=92
=
CK
10
0
K
AC
, 8
byt
e
=1
20
s d
ata
seq=92 timeout interval
ACK=
Time
Figure 3.35
Segment 100 not retransmitted
120
Time
245
246
CHAPTER 3
•
TRANSPORT LAYER
Host A
Host B
Seq=9
2, 8
bytes
Seq=1
X
data
ACK=
00,
20 by
tes
100
data
(loss)
Seq=92 timeout interval
ACK=
Time
Figure 3.36
120
Time
A cumulative acknowledgment avoids retransmission of the
first segment
Doubling the Timeout Interval
We now discuss a few modifications that most TCP implementations employ. The
first concerns the length of the timeout interval after a timer expiration. In this modification, whenever the timeout event occurs, TCP retransmits the not-yetacknowledged segment with the smallest sequence number, as described above. But
each time TCP retransmits, it sets the next timeout interval to twice the previous
value, rather than deriving it from the last EstimatedRTT and DevRTT (as
described in Section 3.5.3). For example, suppose TimeoutInterval associated
with the oldest not yet acknowledged segment is .75 sec when the timer first expires.
TCP will then retransmit this segment and set the new expiration time to 1.5 sec. If
the timer expires again 1.5 sec later, TCP will again retransmit this segment, now
setting the expiration time to 3.0 sec. Thus the intervals grow exponentially after
each retransmission. However, whenever the timer is started after either of the two
other events (that is, data received from application above, and ACK received), the
3.5
•
CONNECTION-ORIENTED TRANSPORT: TCP
247
TimeoutInterval is derived from the most recent values of EstimatedRTT
and DevRTT.
This modification provides a limited form of congestion control. (More comprehensive forms of TCP congestion control will be studied in Section 3.7.) The
timer expiration is most likely caused by congestion in the network, that is, too
many packets arriving at one (or more) router queues in the path between the source
and destination, causing packets to be dropped and/or long queuing delays. In times
of congestion, if the sources continue to retransmit packets persistently, the congestion may get worse. Instead, TCP acts more politely, with each sender retransmitting
after longer and longer intervals. We will see that a similar idea is used by Ethernet
when we study CSMA/CD in Chapter 5.
Fast Retransmit
One of the problems with timeout-triggered retransmissions is that the timeout
period can be relatively long. When a segment is lost, this long timeout period
forces the sender to delay resending the lost packet, thereby increasing the end-toend delay. Fortunately, the sender can often detect packet loss well before the timeout event occurs by noting so-called duplicate ACKs. A duplicate ACK is an ACK
that reacknowledges a segment for which the sender has already received an earlier
acknowledgment. To understand the sender’s response to a duplicate ACK, we must
look at why the receiver sends a duplicate ACK in the first place. Table 3.2 summarizes the TCP receiver’s ACK generation policy [RFC 5681]. When a TCP receiver
receives a segment with a sequence number that is larger than the next, expected,
in-order sequence number, it detects a gap in the data stream—that is, a missing segment. This gap could be the result of lost or reordered segments within the network.
Event
TCP Receiver Action
Arrival of in-order segment with expected sequence number. All
data up to expected sequence number already acknowledged.
Delayed ACK. Wait up to 500 msec for arrival of another in-order segment. If next in-order segment does not arrive in this interval, send an ACK.
Arrival of in-order segment with expected sequence number. One
other in-order segment waiting for ACK transmission.
Immediately send single cumulative ACK, ACKing both in-order segments.
Arrival of out-of-order segment with higher-than-expected sequence
number. Gap detected.
Immediately send duplicate ACK, indicating sequence number of next
expected byte (which is the lower end of the gap).
Arrival of segment that partially or completely fills in gap in
received data.
Immediately send ACK, provided that segment starts at the lower end
of gap.
Table 3.2
TCP ACK Generation Recommendation [RFC 5681]
248
CHAPTER 3
•
TRANSPORT LAYER
Since TCP does not use negative acknowledgments, the receiver cannot send an
explicit negative acknowledgment back to the sender. Instead, it simply reacknowledges (that is, generates a duplicate ACK for) the last in-order byte of data it has
received. (Note that Table 3.2 allows for the case that the receiver does not discard
out-of-order segments.)
Because a sender often sends a large number of segments back to back, if one segment is lost, there will likely be many back-to-back duplicate ACKs. If the TCP sender
receives three duplicate ACKs for the same data, it takes this as an indication that the
segment following the segment that has been ACKed three times has been lost. (In the
homework problems, we consider the question of why the sender waits for three duplicate ACKs, rather than just a single duplicate ACK.) In the case that three duplicate
ACKs are received, the TCP sender performs a fast retransmit [RFC 5681], retransmitting the missing segment before that segment’s timer expires. This is shown in
Figure 3.37, where the second segment is lost, then retransmitted before its timer
expires. For TCP with fast retransmit, the following code snippet replaces the ACK
received event in Figure 3.33:
event: ACK received, with ACK field value of y
if (y > SendBase) {
SendBase=y
if (there are currently any not yet
acknowledged segments)
start timer
}
else { /* a duplicate ACK for already ACKed
segment */
increment number of duplicate ACKs
received for y
if (number of duplicate ACKS received
for y==3)
/* TCP fast retransmit */
resend segment with sequence number y
}
break;
We noted earlier that many subtle issues arise when a timeout/retransmit mechanism is implemented in an actual protocol such as TCP. The procedures above,
which have evolved as a result of more than 20 years of experience with TCP timers,
should convince you that this is indeed the case!
Go-Back-N or Selective Repeat?
Let us close our study of TCP’s error-recovery mechanism by considering the following question: Is TCP a GBN or an SR protocol? Recall that TCP acknowledgments are
cumulative and correctly received but out-of-order segments are not individually
3.5
CONNECTION-ORIENTED TRANSPORT: TCP
•
Host A
Host B
seq=
92,
8 by
seq=
tes
100,
of d
ata
20 b
seq=
ytes
120,
o
f
1
data
5 by
seq=
tes
135,
o
f
d
6 by
ata
seq=
tes
141,
of d
ata
16 b
ytes
of d
ata
X
ack=100
ack=100
ack=100
ack=100
seq
Timeout
Time
Figure 3.37
=10
0,
20
byt
es
of
dat
a
Time
Fast retransmit: retransmitting the missing segment before
the segment’s timer expires
ACKed by the receiver. Consequently, as shown in Figure 3.33 (see also Figure 3.19),
the TCP sender need only maintain the smallest sequence number of a transmitted but
unacknowledged byte (SendBase) and the sequence number of the next byte to be
sent (NextSeqNum). In this sense, TCP looks a lot like a GBN-style protocol. But
there are some striking differences between TCP and Go-Back-N. Many TCP implementations will buffer correctly received but out-of-order segments [Stevens 1994].
Consider also what happens when the sender sends a sequence of segments 1, 2, . . . ,
N, and all of the segments arrive in order without error at the receiver. Further suppose
that the acknowledgment for packet n < N gets lost, but the remaining N – 1 acknowledgments arrive at the sender before their respective timeouts. In this example, GBN
would retransmit not only packet n, but also all of the subsequent packets n + 1, n + 2,
. . . , N. TCP, on the other hand, would retransmit at most one segment, namely, segment n. Moreover, TCP would not even retransmit segment n if the acknowledgment
for segment n + 1 arrived before the timeout for segment n.
249
250
CHAPTER 3
•
TRANSPORT LAYER
A proposed modification to TCP, the so-called selective acknowledgment
[RFC 2018], allows a TCP receiver to acknowledge out-of-order segments selectively rather than just cumulatively acknowledging the last correctly received, inorder segment. When combined with selective retransmission—skipping the
retransmission of segments that have already been selectively acknowledged by the
receiver—TCP looks a lot like our generic SR protocol. Thus, TCP’s error-recovery
mechanism is probably best categorized as a hybrid of GBN and SR protocols.
3.5.5 Flow Control
Recall that the hosts on each side of a TCP connection set aside a receive buffer for
the connection. When the TCP connection receives bytes that are correct and in
sequence, it places the data in the receive buffer. The associated application process
will read data from this buffer, but not necessarily at the instant the data arrives.
Indeed, the receiving application may be busy with some other task and may not
even attempt to read the data until long after it has arrived. If the application is relatively slow at reading the data, the sender can very easily overflow the connection’s
receive buffer by sending too much data too quickly.
TCP provides a flow-control service to its applications to eliminate the possibility
of the sender overflowing the receiver’s buffer. Flow control is thus a speed-matching
service—matching the rate at which the sender is sending against the rate at which the
receiving application is reading. As noted earlier, a TCP sender can also be throttled
due to congestion within the IP network; this form of sender control is referred to as
congestion control, a topic we will explore in detail in Sections 3.6 and 3.7. Even
though the actions taken by flow and congestion control are similar (the throttling of
the sender), they are obviously taken for very different reasons. Unfortunately, many
authors use the terms interchangeably, and the savvy reader would be wise to distinguish between them. Let’s now discuss how TCP provides its flow-control service. In
order to see the forest for the trees, we suppose throughout this section that the TCP
implementation is such that the TCP receiver discards out-of-order segments.
TCP provides flow control by having the sender maintain a variable called the
receive window. Informally, the receive window is used to give the sender an idea of
how much free buffer space is available at the receiver. Because TCP is full-duplex, the
sender at each side of the connection maintains a distinct receive window. Let’s investigate the receive window in the context of a file transfer. Suppose that Host A is sending
a large file to Host B over a TCP connection. Host B allocates a receive buffer to this
connection; denote its size by RcvBuffer. From time to time, the application process
in Host B reads from the buffer. Define the following variables:
• LastByteRead: the number of the last byte in the data stream read from the
buffer by the application process in B
• LastByteRcvd: the number of the last byte in the data stream that has arrived
from the network and has been placed in the receive buffer at B
3.5
•
CONNECTION-ORIENTED TRANSPORT: TCP
RcvBuffer
rwnd
Data
from IP
Spare room
Figure 3.38
TCP data
in buffer
Application
process
The receive window (rwnd) and the receive buffer
(RcvBuffer)
Because TCP is not permitted to overflow the allocated buffer, we must have
LastByteRcvd – LastByteRead ⱕ RcvBuffer
The receive window, denoted rwnd is set to the amount of spare room in the buffer:
rwnd = RcvBuffer – [LastByteRcvd – LastByteRead]
Because the spare room changes with time, rwnd is dynamic. The variable rwnd is
illustrated in Figure 3.38.
How does the connection use the variable rwnd to provide the flow-control
service? Host B tells Host A how much spare room it has in the connection buffer
by placing its current value of rwnd in the receive window field of every segment it
sends to A. Initially, Host B sets rwnd = RcvBuffer. Note that to pull this off,
Host B must keep track of several connection-specific variables.
Host A in turn keeps track of two variables, LastByteSent and LastByteAcked, which have obvious meanings. Note that the difference between these
two variables, LastByteSent – LastByteAcked, is the amount of unacknowledged data that A has sent into the connection. By keeping the amount of
unacknowledged data less than the value of rwnd, Host A is assured that it is not
overflowing the receive buffer at Host B. Thus, Host A makes sure throughout the
connection’s life that
LastByteSent – LastByteAcked ⱕ rwnd
251
252
CHAPTER 3
•
TRANSPORT LAYER
There is one minor technical problem with this scheme. To see this, suppose
Host B’s receive buffer becomes full so that rwnd = 0. After advertising rwnd = 0
to Host A, also suppose that B has nothing to send to A. Now consider what happens. As the application process at B empties the buffer, TCP does not send new segments with new rwnd values to Host A; indeed, TCP sends a segment to Host A
only if it has data to send or if it has an acknowledgment to send. Therefore, Host A
is never informed that some space has opened up in Host B’s receive buffer—Host
A is blocked and can transmit no more data! To solve this problem, the TCP specification requires Host A to continue to send segments with one data byte when B’s
receive window is zero. These segments will be acknowledged by the receiver.
Eventually the buffer will begin to empty and the acknowledgments will contain a
nonzero rwnd value.
The online site at http://www.awl.com/kurose-ross for this book provides an
interactive Java applet that illustrates the operation of the TCP receive window.
Having described TCP’s flow-control service, we briefly mention here that UDP
does not provide flow control. To understand the issue, consider sending a series of
UDP segments from a process on Host A to a process on Host B. For a typical UDP
implementation, UDP will append the segments in a finite-sized buffer that “precedes”
the corresponding socket (that is, the door to the process). The process reads one entire
segment at a time from the buffer. If the process does not read the segments fast
enough from the buffer, the buffer will overflow and segments will get dropped.
3.5.6 TCP Connection Management
In this subsection we take a closer look at how a TCP connection is established and
torn down. Although this topic may not seem particularly thrilling, it is important
because TCP connection establishment can significantly add to perceived delays
(for example, when surfing the Web). Furthermore, many of the most common network attacks—including the incredibly popular SYN flood attack—exploit vulnerabilities in TCP connection management. Let’s first take a look at how a TCP
connection is established. Suppose a process running in one host (client) wants to
initiate a connection with another process in another host (server). The client application process first informs the client TCP that it wants to establish a connection to
a process in the server. The TCP in the client then proceeds to establish a TCP connection with the TCP in the server in the following manner:
• Step 1. The client-side TCP first sends a special TCP segment to the server-side
TCP. This special segment contains no application-layer data. But one of the flag
bits in the segment’s header (see Figure 3.29), the SYN bit, is set to 1. For this
reason, this special segment is referred to as a SYN segment. In addition, the
client randomly chooses an initial sequence number (client_isn) and puts
this number in the sequence number field of the initial TCP SYN segment. This
segment is encapsulated within an IP datagram and sent to the server. There has
3.5
•
CONNECTION-ORIENTED TRANSPORT: TCP
been considerable interest in properly randomizing the choice of the
client_isn in order to avoid certain security attacks [CERT 2001–09].
• Step 2. Once the IP datagram containing the TCP SYN segment arrives at the
server host (assuming it does arrive!), the server extracts the TCP SYN segment
from the datagram, allocates the TCP buffers and variables to the connection, and
sends a connection-granted segment to the client TCP. (We’ll see in Chapter 8 that
the allocation of these buffers and variables before completing the third step of the
three-way handshake makes TCP vulnerable to a denial-of-service attack known
as SYN flooding.) This connection-granted segment also contains no applicationlayer data. However, it does contain three important pieces of information in the
segment header. First, the SYN bit is set to 1. Second, the acknowledgment field
of the TCP segment header is set to client_isn+1. Finally, the server
chooses its own initial sequence number (server_isn) and puts this value in
the sequence number field of the TCP segment header. This connection-granted
segment is saying, in effect, “I received your SYN packet to start a connection
with your initial sequence number, client_isn. I agree to establish this connection. My own initial sequence number is server_isn.” The connectiongranted segment is referred to as a SYNACK segment.
• Step 3. Upon receiving the SYNACK segment, the client also allocates buffers
and variables to the connection. The client host then sends the server yet another
segment; this last segment acknowledges the server’s connection-granted segment (the client does so by putting the value server_isn+1 in the acknowledgment field of the TCP segment header). The SYN bit is set to zero, since the
connection is established. This third stage of the three-way handshake may carry
client-to-server data in the segment payload.
Once these three steps have been completed, the client and server hosts can send
segments containing data to each other. In each of these future segments, the SYN bit
will be set to zero. Note that in order to establish the connection, three packets are sent
between the two hosts, as illustrated in Figure 3.39. For this reason, this connectionestablishment procedure is often referred to as a three-way handshake. Several
aspects of the TCP three-way handshake are explored in the homework problems
(Why are initial sequence numbers needed? Why is a three-way handshake, as
opposed to a two-way handshake, needed?). It’s interesting to note that a rock climber
and a belayer (who is stationed below the rock climber and whose job it is to handle
the climber’s safety rope) use a three-way-handshake communication protocol that is
identical to TCP’s to ensure that both sides are ready before the climber begins ascent.
All good things must come to an end, and the same is true with a TCP connection. Either of the two processes participating in a TCP connection can end the connection. When a connection ends, the “resources” (that is, the buffers and variables)
in the hosts are deallocated. As an example, suppose the client decides to close the
connection, as shown in Figure 3.40. The client application process issues a close
253
254
CHAPTER 3
•
TRANSPORT LAYER
Client host
Connection
request
SYN=1
, seq
=cli
ent_i
sn
,
isn
er_
v
r
se
+1
eq=
, s t_isn
1
=
SYN clien
=
ack
SYN
=
ack 0, se
=se
q
rve =clie
r_i
n
sn+ t_isn
1
+1,
ACK
Time
Figure 3.39
Server host
Connection
granted
Time
TCP three-way handshake: segment exchange
command. This causes the client TCP to send a special TCP segment to the server
process. This special segment has a flag bit in the segment’s header, the FIN bit
(see Figure 3.29), set to 1. When the server receives this segment, it sends the client
an acknowledgment segment in return. The server then sends its own shutdown
segment, which has the FIN bit set to 1. Finally, the client acknowledges the
server’s shutdown segment. At this point, all the resources in the two hosts are now
deallocated.
During the life of a TCP connection, the TCP protocol running in each host
makes transitions through various TCP states. Figure 3.41 illustrates a typical
sequence of TCP states that are visited by the client TCP. The client TCP begins in
the CLOSED state. The application on the client side initiates a new TCP connection (by creating a Socket object in our Java examples as in the Python examples
from Chapter 2). This causes TCP in the client to send a SYN segment to TCP in the
server. After having sent the SYN segment, the client TCP enters the SYN_SENT
state. While in the SYN_SENT state, the client TCP waits for a segment from the
server TCP that includes an acknowledgment for the client’s previous segment and
has the SYN bit set to 1. Having received such a segment, the client TCP enters the
ESTABLISHED state. While in the ESTABLISHED state, the TCP client can send
and receive TCP segments containing payload (that is, application-generated) data.
3.5
Client
Close
•
CONNECTION-ORIENTED TRANSPORT: TCP
Server
FIN
ACK
FIN
Close
ACK
Timed wait
Closed
Time
Figure 3.40
Time
Closing a TCP connection
Suppose that the client application decides it wants to close the connection.
(Note that the server could also choose to close the connection.) This causes the
client TCP to send a TCP segment with the FIN bit set to 1 and to enter the
FIN_WAIT_1 state. While in the FIN_WAIT_1 state, the client TCP waits for a TCP
segment from the server with an acknowledgment. When it receives this segment,
the client TCP enters the FIN_WAIT_2 state. While in the FIN_WAIT_2 state, the
client waits for another segment from the server with the FIN bit set to 1; after
receiving this segment, the client TCP acknowledges the server’s segment and
enters the TIME_WAIT state. The TIME_WAIT state lets the TCP client resend the
final acknowledgment in case the ACK is lost. The time spent in the TIME_WAIT
state is implementation-dependent, but typical values are 30 seconds, 1 minute, and
2 minutes. After the wait, the connection formally closes and all resources on the
client side (including port numbers) are released.
Figure 3.42 illustrates the series of states typically visited by the server-side
TCP, assuming the client begins connection teardown. The transitions are selfexplanatory. In these two state-transition diagrams, we have only shown how a TCP
connection is normally established and shut down. We have not described what
255
256
CHAPTER 3
•
TRANSPORT LAYER
Client application
initiates a TCP connection
CLOSED
Wait 30 seconds
Send SYN
TIME_WAIT
SYN_SENT
Receive SYN & ACK,
send ACK
Receive FIN,
send ACK
FIN_WAIT_2
ESTABLISHED
Send FIN
Receive ACK,
send nothing
Figure 3.41
FIN_WAIT_1
Client application
initiates close connection
A typical sequence of TCP states visited by a client TCP
Receive ACK,
send nothing
CLOSED
LAST_ACK
Server application
creates a listen socket
LISTEN
Receive SYN
send SYN & ACK
Send FIN
CLOSE_WAIT
SYN_RCVD
Receive ACK,
send nothing
Receive FIN,
send ACK
ESTABLISHED
Figure 3.42
A typical sequence of TCP states visited by a server-side TCP
3.5
•
CONNECTION-ORIENTED TRANSPORT: TCP
FOCUS ON SECURITY
THE SYN FLOOD ATTACK
We’ve seen in our discussion of TCP’s three-way handshake that a server allocates
and initializes connection variables and buffers in response to a received SYN. The
server then sends a SYNACK in response, and awaits an ACK segment from the
client. If the client does not send an ACK to complete the third step of this 3-way
handshake, eventually (often after a minute or more) the server will terminate the halfopen connection and reclaim the allocated resources.
This TCP connection management protocol sets the stage for a classic Denial of
Service (DoS) attack known as the SYN flood attack. In this attack, the attacker(s)
send a large number of TCP SYN segments, without completing the third handshake
step. With this deluge of SYN segments, the server’s connection resources become
exhausted as they are allocated (but never used!) for half-open connections; legitimate clients are then denied service. Such SYN flooding attacks were among the first
documented DoS attacks [CERT SYN 1996]. Fortunately, an effective defense known
as SYN cookies [RFC 4987] are now deployed in most major operating systems.
SYN cookies work as follows:
o When the server receives a SYN segment, it does not know if the segment is coming from a legitimate user or is part of a SYN flood attack. So, instead of creating
a half-open TCP connection for this SYN, the server creates an initial TCP
sequence number that is a complicated function (hash function) of source and destination IP addresses and port numbers of the SYN segment, as well as a secret
number only known to the server. This carefully crafted initial sequence number is
the so-called “cookie.” The server then sends the client a SYNACK packet with this
special initial sequence number. Importantly, the server does not remember the
cookie or any other state information corresponding to the SYN.
o A legitimate client will return an ACK segment. When the server receives this
ACK, it must verify that the ACK corresponds to some SYN sent earlier. But how is
this done if the server maintains no memory about SYN segments? As you may
have guessed, it is done with the cookie. Recall that for a legitimate ACK, the
value in the acknowledgment field is equal to the initial sequence number in the
SYNACK (the cookie value in this case) plus one (see Figure 3.39). The server can
then run the same hash function using the source and destination IP address and
port numbers in the SYNACK (which are the same as in the original SYN) and the
secret number. If the result of the function plus one is the same as the acknowledgment (cookie) value in the client’s SYNACK, the server concludes that the ACK corresponds to an earlier SYN segment and is hence valid. The server then creates a
fully open connection along with a socket.
o On the other hand, if the client does not return an ACK segment, then the original
SYN has done no harm at the server, since the server hasn’t yet allocated any
resources in response to the original bogus SYN.
257
258
CHAPTER 3
•
TRANSPORT LAYER
happens in certain pathological scenarios, for example, when both sides of a connection want to initiate or shut down at the same time. If you are interested in learning about this and other advanced issues concerning TCP, you are encouraged to see
Stevens’ comprehensive book [Stevens 1994].
Our discussion above has assumed that both the client and server are prepared
to communicate, i.e., that the server is listening on the port to which the client sends
its SYN segment. Let’s consider what happens when a host receives a TCP segment
whose port numbers or source IP address do not match with any of the ongoing
sockets in the host. For example, suppose a host receives a TCP SYN packet with
destination port 80, but the host is not accepting connections on port 80 (that is, it is
not running a Web server on port 80). Then the host will send a special reset segment to the source. This TCP segment has the RST flag bit (see Section 3.5.2) set to
1. Thus, when a host sends a reset segment, it is telling the source “I don’t have a
socket for that segment. Please do not resend the segment.” When a host receives a
UDP packet whose destination port number doesn’t match with an ongoing UDP
socket, the host sends a special ICMP datagram, as discussed in Chapter 4.
Now that we have a good understanding of TCP connection management, let’s
revisit the nmap port-scanning tool and examine more closely how it works. To explore
a specific TCP port, say port 6789, on a target host, nmap will send a TCP SYN segment with destination port 6789 to that host. There are three possible outcomes:
• The source host receives a TCP SYNACK segment from the target host. Since this
means that an application is running with TCP port 6789 on the target post, nmap
returns “open.”
• The source host receives a TCP RST segment from the target host. This means that
the SYN segment reached the target host, but the target host is not running an application with TCP port 6789. But the attacker at least knows that the segments destined to the host at port 6789 are not blocked by any firewall on the path between
source and target hosts. (Firewalls are discussed in Chapter 8.)
• The source receives nothing. This likely means that the SYN segment was blocked
by an intervening firewall and never reached the target host.
Nmap is a powerful tool, which can “case the joint” not only for open TCP ports,
but also for open UDP ports, for firewalls and their configurations, and even for the versions of applications and operating systems. Most of this is done by manipulating TCP
connection-management segments [Skoudis 2006]. You can download nmap from
www.nmap.org.
This completes our introduction to error control and flow control in TCP. In
Section 3.7 we’ll return to TCP and look at TCP congestion control in some depth.
Before doing so, however, we first step back and examine congestion-control issues
in a broader context.
3.6
•
PRINCIPLES OF CONGESTION CONTROL
3.6 Principles of Congestion Control
In the previous sections, we examined both the general principles and specific
TCP mechanisms used to provide for a reliable data transfer service in the face of
packet loss. We mentioned earlier that, in practice, such loss typically results from
the overflowing of router buffers as the network becomes congested. Packet
retransmission thus treats a symptom of network congestion (the loss of a specific
transport-layer segment) but does not treat the cause of network congestion—too
many sources attempting to send data at too high a rate. To treat the cause of network congestion, mechanisms are needed to throttle senders in the face of network
congestion.
In this section, we consider the problem of congestion control in a general context, seeking to understand why congestion is a bad thing, how network congestion
is manifested in the performance received by upper-layer applications, and various
approaches that can be taken to avoid, or react to, network congestion. This more
general study of congestion control is appropriate since, as with reliable data transfer, it is high on our “top-ten” list of fundamentally important problems in networking. We conclude this section with a discussion of congestion control in the
available bit-rate (ABR) service in asynchronous transfer mode (ATM)
networks. The following section contains a detailed study of TCP’s congestioncontrol algorithm.
3.6.1 The Causes and the Costs of Congestion
Let’s begin our general study of congestion control by examining three increasingly
complex scenarios in which congestion occurs. In each case, we’ll look at why congestion occurs in the first place and at the cost of congestion (in terms of resources
not fully utilized and poor performance received by the end systems). We’ll not (yet)
focus on how to react to, or avoid, congestion but rather focus on the simpler issue
of understanding what happens as hosts increase their transmission rate and the network becomes congested.
Scenario 1: Two Senders, a Router with Infinite Buffers
We begin by considering perhaps the simplest congestion scenario possible: Two
hosts (A and B) each have a connection that shares a single hop between source and
destination, as shown in Figure 3.43.
Let’s assume that the application in Host A is sending data into the connection
(for example, passing data to the transport-level protocol via a socket) at an average rate of ␭in bytes/sec. These data are original in the sense that each unit of data
is sent into the socket only once. The underlying transport-level protocol is a
259
260
CHAPTER 3
•
TRANSPORT LAYER
λout
λin: original data
Host A
Host B
Host C
Host D
Unlimited shared
output link buffers
Figure 3.43
Congestion scenario 1: Two connections sharing a single
hop with infinite buffers
simple one. Data is encapsulated and sent; no error recovery (for example, retransmission), flow control, or congestion control is performed. Ignoring the additional
overhead due to adding transport- and lower-layer header information, the rate at
which Host A offers traffic to the router in this first scenario is thus ␭in bytes/sec.
Host B operates in a similar manner, and we assume for simplicity that it too is
sending at a rate of ␭in bytes/sec. Packets from Hosts A and B pass through a
router and over a shared outgoing link of capacity R. The router has buffers that
allow it to store incoming packets when the packet-arrival rate exceeds the outgoing link’s capacity. In this first scenario, we assume that the router has an infinite
amount of buffer space.
Figure 3.44 plots the performance of Host A’s connection under this first
scenario. The left graph plots the per-connection throughput (number of bytes per
second at the receiver) as a function of the connection-sending rate. For a sending
rate between 0 and R/2, the throughput at the receiver equals the sender’s sending
rate—everything sent by the sender is received at the receiver with a finite delay.
When the sending rate is above R/2, however, the throughput is only R/2. This upper
limit on throughput is a consequence of the sharing of link capacity between two
connections. The link simply cannot deliver packets to a receiver at a steady-state
rate that exceeds R/2. No matter how high Hosts A and B set their sending rates,
they will each never see a throughput higher than R/2.
Achieving a per-connection throughput of R/2 might actually appear to be a
good thing, because the link is fully utilized in delivering packets to their destinations. The right-hand graph in Figure 3.44, however, shows the consequence of
operating near link capacity. As the sending rate approaches R/2 (from the left), the
average delay becomes larger and larger. When the sending rate exceeds R/2, the
3.6
•
PRINCIPLES OF CONGESTION CONTROL
λout
Delay
R/2
λ in
a.
Figure 3.44
R/2
λ in
R/2
b.
Congestion scenario 1: Throughput and delay as a function
of host sending rate
average number of queued packets in the router is unbounded, and the average delay
between source and destination becomes infinite (assuming that the connections
operate at these sending rates for an infinite period of time and there is an infinite
amount of buffering available). Thus, while operating at an aggregate throughput of
near R may be ideal from a throughput standpoint, it is far from ideal from a delay
standpoint. Even in this (extremely) idealized scenario, we’ve already found one
cost of a congested network—large queuing delays are experienced as the packetarrival rate nears the link capacity.
Scenario 2: Two Senders and a Router with Finite Buffers
Let us now slightly modify scenario 1 in the following two ways (see Figure 3.45).
First, the amount of router buffering is assumed to be finite. A consequence of this
real-world assumption is that packets will be dropped when arriving to an alreadyfull buffer. Second, we assume that each connection is reliable. If a packet containing a transport-level segment is dropped at the router, the sender will eventually
retransmit it. Because packets can be retransmitted, we must now be more careful
with our use of the term sending rate. Specifically, let us again denote the rate at
which the application sends original data into the socket by ␭in bytes/sec. The rate at
which the transport layer sends segments (containing original data and retransmitted data) into the network will be denoted ␭⬘in bytes/sec. ␭⬘in is sometimes referred to
as the offered load to the network.
The performance realized under scenario 2 will now depend strongly on
how retransmission is performed. First, consider the unrealistic case that Host A is
able to somehow (magically!) determine whether or not a buffer is free in the router
and thus sends a packet only when a buffer is free. In this case, no loss would occur,
261
262
CHAPTER 3
•
TRANSPORT LAYER
λin: original data
λ’in: original data, plus
retransmitted data
Host A
Host B
λout
Host C
Host D
Finite shared output
link buffers
Figure 3.45
Scenario 2: Two hosts (with retransmissions) and a router
with finite buffers
␭in would be equal to ␭⬘in, and the throughput of the connection would be equal to
␭in. This case is shown in Figure 3.46(a). From a throughput standpoint, performance is ideal—everything that is sent is received. Note that the average host sending
rate cannot exceed R/2 under this scenario, since packet loss is assumed never
to occur.
Consider next the slightly more realistic case that the sender retransmits only
when a packet is known for certain to be lost. (Again, this assumption is a bit of a
stretch. However, it is possible that the sending host might set its timeout large
enough to be virtually assured that a packet that has not been acknowledged has
been lost.) In this case, the performance might look something like that shown in
Figure 3.46(b). To appreciate what is happening here, consider the case that the
offered load, ␭⬘in (the rate of original data transmission plus retransmissions), equals
R/2. According to Figure 3.46(b), at this value of the offered load, the rate at which
data are delivered to the receiver application is R/3. Thus, out of the 0.5R units of
data transmitted, 0.333R bytes/sec (on average) are original data and 0.166R bytes/
sec (on average) are retransmitted data. We see here another cost of a congested network—the sender must perform retransmissions in order to compensate for dropped
(lost) packets due to buffer overflow.
Finally, let us consider the case that the sender may time out prematurely and
retransmit a packet that has been delayed in the queue but not yet lost. In this case,
both the original data packet and the retransmission may reach the receiver. Of
3.6
•
PRINCIPLES OF CONGESTION CONTROL
R/2
R/2
263
R/2
λ’in
a.
Figure 3.46
λout
λout
λout
R/3
R/2
λ’in
R/2
b.
R/4
λ’in
c.
Scenario 2 performance with finite buffers
course, the receiver needs but one copy of this packet and will discard the retransmission. In this case, the work done by the router in forwarding the retransmitted
copy of the original packet was wasted, as the receiver will have already received
the original copy of this packet. The router would have better used the link transmission capacity to send a different packet instead. Here then is yet another cost of
a congested network—unneeded retransmissions by the sender in the face of large
delays may cause a router to use its link bandwidth to forward unneeded copies of a
packet. Figure 3.46 (c) shows the throughput versus offered load when each packet
is assumed to be forwarded (on average) twice by the router. Since each packet is
forwarded twice, the throughput will have an asymptotic value of R/4 as the offered
load approaches R/ 2.
Scenario 3: Four Senders, Routers with Finite Buffers, and
Multihop Paths
In our final congestion scenario, four hosts transmit packets, each over overlapping two-hop paths, as shown in Figure 3.47. We again assume that each host
uses a timeout/retransmission mechanism to implement a reliable data transfer
service, that all hosts have the same value of ␭ in , and that all router links have
capacity R bytes/sec.
Let’s consider the connection from Host A to Host C, passing through routers
R1 and R2. The A–C connection shares router R1 with the D–B connection and
shares router R2 with the B–D connection. For extremely small values of ␭in, buffer
overflows are rare (as in congestion scenarios 1 and 2), and the throughput approximately equals the offered load. For slightly larger values of ␭in, the corresponding
throughput is also larger, since more original data is being transmitted into the
R/2
264
CHAPTER 3
•
TRANSPORT LAYER
λin: original data
λ’in : original
data, plus
retransmitted
data
Host A
λout
Host B
R1
R4
R2
Finite shared output
link buffers
Host D
Host C
R3
Figure 3.47
Four senders, routers with finite buffers, and multihop paths
network and delivered to the destination, and overflows are still rare. Thus, for small
values of ␭in, an increase in ␭in results in an increase in ␭out.
Having considered the case of extremely low traffic, let’s next examine the
case that ␭in (and hence ␭⬘in ) is extremely large. Consider router R2. The A–C
traffic arriving to router R2 (which arrives at R2 after being forwarded from R1)
can have an arrival rate at R2 that is at most R, the capacity of the link from R1
to R2, regardless of the value of ␭in. If ␭⬘in is extremely large for all connections
(including the B–D connection), then the arrival rate of B–D traffic at R2 can be
much larger than that of the A–C traffic. Because the A–C and B–D traffic must
compete at router R2 for the limited amount of buffer space, the amount of A–C
traffic that successfully gets through R2 (that is, is not lost due to buffer overflow) becomes smaller and smaller as the offered load from B–D gets larger and
larger. In the limit, as the offered load approaches infinity, an empty buffer at R2
3.6
•
PRINCIPLES OF CONGESTION CONTROL
is immediately filled by a B–D packet, and the throughput of the A–C connection
at R2 goes to zero. This, in turn, implies that the A–C end-to-end throughput goes
to zero in the limit of heavy traffic. These considerations give rise to the offered
load versus throughput tradeoff shown in Figure 3.48.
The reason for the eventual decrease in throughput with increasing offered
load is evident when one considers the amount of wasted work done by the network. In the high-traffic scenario outlined above, whenever a packet is dropped
at a second-hop router, the work done by the first-hop router in forwarding a
packet to the second-hop router ends up being “wasted.” The network would
have been equally well off (more accurately, equally bad off) if the first router
had simply discarded that packet and remained idle. More to the point, the transmission capacity used at the first router to forward the packet to the second router
could have been much more profitably used to transmit a different packet. (For
example, when selecting a packet for transmission, it might be better for a router
to give priority to packets that have already traversed some number of upstream
routers.) So here we see yet another cost of dropping a packet due to congestion—when a packet is dropped along a path, the transmission capacity that was
used at each of the upstream links to forward that packet to the point at which it
is dropped ends up having been wasted.
3.6.2 Approaches to Congestion Control
In Section 3.7, we’ll examine TCP’s specific approach to congestion control in great
detail. Here, we identify the two broad approaches to congestion control that are
taken in practice and discuss specific network architectures and congestion-control
protocols embodying these approaches.
λout
R/2
λ’in
Figure 3.48
Scenario 3 performance with finite buffers and multihop
paths
265
266
CHAPTER 3
•
TRANSPORT LAYER
At the broadest level, we can distinguish among congestion-control approaches
by whether the network layer provides any explicit assistance to the transport layer
for congestion-control purposes:
• End-to-end congestion control. In an end-to-end approach to congestion control,
the network layer provides no explicit support to the transport layer for congestioncontrol purposes. Even the presence of congestion in the network must be inferred
by the end systems based only on observed network behavior (for example, packet
loss and delay). We will see in Section 3.7 that TCP must necessarily take this endto-end approach toward congestion control, since the IP layer provides no feedback
to the end systems regarding network congestion. TCP segment loss (as indicated
by a timeout or a triple duplicate acknowledgment) is taken as an indication of network congestion and TCP decreases its window size accordingly. We will also see
a more recent proposal for TCP congestion control that uses increasing round-trip
delay values as indicators of increased network congestion.
• Network-assisted congestion control. With network-assisted congestion control,
network-layer components (that is, routers) provide explicit feedback to the
sender regarding the congestion state in the network. This feedback may be as
simple as a single bit indicating congestion at a link. This approach was taken in
the early IBM SNA [Schwartz 1982] and DEC DECnet [Jain 1989; Ramakrishnan 1990] architectures, was recently proposed for TCP/IP networks [Floyd TCP
1994; RFC 3168], and is used in ATM available bit-rate (ABR) congestion control as well, as discussed below. More sophisticated network feedback is also possible. For example, one form of ATM ABR congestion control that we will study
shortly allows a router to inform the sender explicitly of the transmission rate it
(the router) can support on an outgoing link. The XCP protocol [Katabi 2002] provides router-computed feedback to each source, carried in the packet header,
regarding how that source should increase or decrease its transmission rate.
For network-assisted congestion control, congestion information is typically fed
back from the network to the sender in one of two ways, as shown in Figure 3.49.
Direct feedback may be sent from a network router to the sender. This form of notification typically takes the form of a choke packet (essentially saying, “I’m congested!”). The second form of notification occurs when a router marks/updates a field
in a packet flowing from sender to receiver to indicate congestion. Upon receipt of a
marked packet, the receiver then notifies the sender of the congestion indication.
Note that this latter form of notification takes at least a full round-trip time.
3.6.3 Network-Assisted Congestion-Control Example:
ATM ABR Congestion Control
We conclude this section with a brief case study of the congestion-control algorithm
in ATM ABR—a protocol that takes a network-assisted approach toward congestion
control. We stress that our goal here is not to describe aspects of the ATM architecture
3.6
•
PRINCIPLES OF CONGESTION CONTROL
Host A
Host B
Network feedback via receiver
Direct network
feedback
Figure 3.49
Two feedback pathways for network-indicated congestion
information
in great detail, but rather to illustrate a protocol that takes a markedly different
approach toward congestion control from that of the Internet’s TCP protocol. Indeed,
we only present below those few aspects of the ATM architecture that are needed to
understand ABR congestion control.
Fundamentally ATM takes a virtual-circuit (VC) oriented approach toward
packet switching. Recall from our discussion in Chapter 1, this means that each
switch on the source-to-destination path will maintain state about the source-todestination VC. This per-VC state allows a switch to track the behavior of individual senders (e.g., tracking their average transmission rate) and to take
source-specific congestion-control actions (such as explicitly signaling to the
sender to reduce its rate when the switch becomes congested). This per-VC state
at network switches makes ATM ideally suited to perform network-assisted congestion control.
ABR has been designed as an elastic data transfer service in a manner reminiscent of TCP. When the network is underloaded, ABR service should be able to take
advantage of the spare available bandwidth; when the network is congested, ABR
service should throttle its transmission rate to some predetermined minimum transmission rate. A detailed tutorial on ATM ABR congestion control and traffic management is provided in [Jain 1996].
Figure 3.50 shows the framework for ATM ABR congestion control. In our
discussion we adopt ATM terminology (for example, using the term switch rather
than router, and the term cell rather than packet). With ATM ABR service, data
cells are transmitted from a source to a destination through a series of intermediate switches. Interspersed with the data cells are resource-management cells
267
268
CHAPTER 3
•
TRANSPORT LAYER
Source
Destination
Switch
Switch
Key:
RM cells
Data cells
Figure 3.50
Congestion-control framework for ATM ABR service
(RM cells); these RM cells can be used to convey congestion-related information
among the hosts and switches. When an RM cell arrives at a destination, it will
be turned around and sent back to the sender (possibly after the destination has
modified the contents of the RM cell). It is also possible for a switch to generate
an RM cell itself and send this RM cell directly to a source. RM cells can thus be
used to provide both direct network feedback and network feedback via the
receiver, as shown in Figure 3.50.
ATM ABR congestion control is a rate-based approach. That is, the sender
explicitly computes a maximum rate at which it can send and regulates itself accordingly. ABR provides three mechanisms for signaling congestion-related information
from the switches to the receiver:
• EFCI bit. Each data cell contains an explicit forward congestion indication
(EFCI) bit. A congested network switch can set the EFCI bit in a data cell to
1 to signal congestion to the destination host. The destination must check the
EFCI bit in all received data cells. When an RM cell arrives at the destination,
if the most recently received data cell had the EFCI bit set to 1, then the destination sets the congestion indication bit (the CI bit) of the RM cell to 1 and
sends the RM cell back to the sender. Using the EFCI in data cells and the CI
bit in RM cells, a sender can thus be notified about congestion at a network
switch.
• CI and NI bits. As noted above, sender-to-receiver RM cells are interspersed
with data cells. The rate of RM cell interspersion is a tunable parameter, with
the default value being one RM cell every 32 data cells. These RM cells have a
congestion indication (CI) bit and a no increase (NI) bit that can be set by a
3.7
•
TCP CONGESTION CONTROL
congested network switch. Specifically, a switch can set the NI bit in a passing
RM cell to 1 under mild congestion and can set the CI bit to 1 under severe
congestion conditions. When a destination host receives an RM cell, it will
send the RM cell back to the sender with its CI and NI bits intact (except that
CI may be set to 1 by the destination as a result of the EFCI mechanism
described above).
• ER setting. Each RM cell also contains a 2-byte explicit rate (ER) field. A congested switch may lower the value contained in the ER field in a passing RM
cell. In this manner, the ER field will be set to the minimum supportable rate of
all switches on the source-to-destination path.
An ATM ABR source adjusts the rate at which it can send cells as a function of
the CI, NI, and ER values in a returned RM cell. The rules for making this rate
adjustment are rather complicated and a bit tedious. The interested reader is referred
to [Jain 1996] for details.
3.7 TCP Congestion Control
In this section we return to our study of TCP. As we learned in Section 3.5, TCP provides a reliable transport service between two processes running on different hosts.
Another key component of TCP is its congestion-control mechanism. As indicated
in the previous section, TCP must use end-to-end congestion control rather than network-assisted congestion control, since the IP layer provides no explicit feedback to
the end systems regarding network congestion.
The approach taken by TCP is to have each sender limit the rate at which it
sends traffic into its connection as a function of perceived network congestion. If a
TCP sender perceives that there is little congestion on the path between itself and
the destination, then the TCP sender increases its send rate; if the sender perceives
that there is congestion along the path, then the sender reduces its send rate. But this
approach raises three questions. First, how does a TCP sender limit the rate at which
it sends traffic into its connection? Second, how does a TCP sender perceive that
there is congestion on the path between itself and the destination? And third, what
algorithm should the sender use to change its send rate as a function of perceived
end-to-end congestion?
Let’s first examine how a TCP sender limits the rate at which it sends traffic
into its connection. In Section 3.5 we saw that each side of a TCP connection consists
of a receive buffer, a send buffer, and several variables (LastByteRead, rwnd,
and so on). The TCP congestion-control mechanism operating at the sender keeps
track of an additional variable, the congestion window. The congestion window,
denoted cwnd, imposes a constraint on the rate at which a TCP sender can send traffic
269
270
CHAPTER 3
•
TRANSPORT LAYER
into the network. Specifically, the amount of unacknowledged data at a sender may
not exceed the minimum of cwnd and rwnd, that is:
LastByteSent – LastByteAcked ⱕ min{cwnd, rwnd}
In order to focus on congestion control (as opposed to flow control), let us henceforth assume that the TCP receive buffer is so large that the receive-window constraint can be ignored; thus, the amount of unacknowledged data at the sender is
solely limited by cwnd. We will also assume that the sender always has data to
send, i.e., that all segments in the congestion window are sent.
The constraint above limits the amount of unacknowledged data at the sender
and therefore indirectly limits the sender’s send rate. To see this, consider a connection for which loss and packet transmission delays are negligible. Then, roughly, at
the beginning of every RTT, the constraint permits the sender to send cwnd bytes of
data into the connection; at the end of the RTT the sender receives acknowledgments for the data. Thus the sender’s send rate is roughly cwnd/RTT bytes/sec. By
adjusting the value of cwnd, the sender can therefore adjust the rate at which it
sends data into its connection.
Let’s next consider how a TCP sender perceives that there is congestion on
the path between itself and the destination. Let us define a “loss event” at a TCP
sender as the occurrence of either a timeout or the receipt of three duplicate
ACKs from the receiver. (Recall our discussion in Section 3.5.4 of the timeout
event in Figure 3.33 and the subsequent modification to include fast retransmit
on receipt of three duplicate ACKs.) When there is excessive congestion, then
one (or more) router buffers along the path overflows, causing a datagram (containing a TCP segment) to be dropped. The dropped datagram, in turn, results in
a loss event at the sender—either a timeout or the receipt of three duplicate
ACKs—which is taken by the sender to be an indication of congestion on the
sender-to-receiver path.
Having considered how congestion is detected, let’s next consider the more
optimistic case when the network is congestion-free, that is, when a loss event
doesn’t occur. In this case, acknowledgments for previously unacknowledged
segments will be received at the TCP sender. As we’ll see, TCP will take the
arrival of these acknowledgments as an indication that all is well—that segments
being transmitted into the network are being successfully delivered to the
destination—and will use acknowledgments to increase its congestion window
size (and hence its transmission rate). Note that if acknowledgments arrive at a
relatively slow rate (e.g., if the end-end path has high delay or contains a
low-bandwidth link), then the congestion window will be increased at a relatively
slow rate. On the other hand, if acknowledgments arrive at a high rate, then the
congestion window will be increased more quickly. Because TCP uses
3.7
•
TCP CONGESTION CONTROL
acknowledgments to trigger (or clock) its increase in congestion window size,
TCP is said to be self-clocking.
Given the mechanism of adjusting the value of cwnd to control the sending rate,
the critical question remains: How should a TCP sender determine the rate at which
it should send? If TCP senders collectively send too fast, they can congest the network, leading to the type of congestion collapse that we saw in Figure 3.48. Indeed,
the version of TCP that we’ll study shortly was developed in response to observed
Internet congestion collapse [Jacobson 1988] under earlier versions of TCP. However, if TCP senders are too cautious and send too slowly, they could under utilize
the bandwidth in the network; that is, the TCP senders could send at a higher rate
without congesting the network. How then do the TCP senders determine their sending rates such that they don’t congest the network but at the same time make use of
all the available bandwidth? Are TCP senders explicitly coordinated, or is there a
distributed approach in which the TCP senders can set their sending rates based only
on local information? TCP answers these questions using the following guiding
principles:
• A lost segment implies congestion, and hence, the TCP sender’s rate should be
decreased when a segment is lost. Recall from our discussion in Section 3.5.4,
that a timeout event or the receipt of four acknowledgments for a given segment (one original ACK and then three duplicate ACKs) is interpreted as an
implicit “loss event” indication of the segment following the quadruply ACKed
segment, triggering a retransmission of the lost segment. From a congestioncontrol standpoint, the question is how the TCP sender should decrease its congestion window size, and hence its sending rate, in response to this inferred
loss event.
• An acknowledged segment indicates that the network is delivering the sender’s
segments to the receiver, and hence, the sender’s rate can be increased when an
ACK arrives for a previously unacknowledged segment. The arrival of acknowledgments is taken as an implicit indication that all is well—segments are being
successfully delivered from sender to receiver, and the network is thus not congested. The congestion window size can thus be increased.
• Bandwidth probing. Given ACKs indicating a congestion-free source-to-destination
path and loss events indicating a congested path, TCP’s strategy for adjusting its
transmission rate is to increase its rate in response to arriving ACKs until a loss
event occurs, at which point, the transmission rate is decreased. The TCP sender
thus increases its transmission rate to probe for the rate that at which congestion
onset begins, backs off from that rate, and then to begins probing again to see if
the congestion onset rate has changed. The TCP sender’s behavior is perhaps analogous to the child who requests (and gets) more and more goodies until finally
he/she is finally told “No!”, backs off a bit, but then begins making requests
271
272
CHAPTER 3
•
TRANSPORT LAYER
again shortly afterwards. Note that there is no explicit signaling of congestion
state by the network—ACKs and loss events serve as implicit signals—and that
each TCP sender acts on local information asynchronously from other TCP
senders.
Given this overview of TCP congestion control, we’re now in a position to consider
the details of the celebrated TCP congestion-control algorithm, which was first
described in [Jacobson 1988] and is standardized in [RFC 5681]. The algorithm has
three major components: (1) slow start, (2) congestion avoidance, and (3) fast recovery. Slow start and congestion avoidance are mandatory components of TCP, differing in how they increase the size of cwnd in response to received ACKs. We’ll see
shortly that slow start increases the size of cwnd more rapidly (despite its name!)
than congestion avoidance. Fast recovery is recommended, but not required, for
TCP senders.
Slow Start
When a TCP connection begins, the value of cwnd is typically initialized to a
small value of 1 MSS [RFC 3390], resulting in an initial sending rate of roughly
MSS/RTT. For example, if MSS = 500 bytes and RTT = 200 msec, the resulting
initial sending rate is only about 20 kbps. Since the available bandwidth to the
TCP sender may be much larger than MSS/RTT, the TCP sender would like to
find the amount of available bandwidth quickly. Thus, in the slow-start state, the
value of cwnd begins at 1 MSS and increases by 1 MSS every time a transmitted
segment is first acknowledged. In the example of Figure 3.51, TCP sends the first
segment into the network and waits for an acknowledgment. When this acknowledgment arrives, the TCP sender increases the congestion window by one MSS
and sends out two maximum-sized segments. These segments are then acknowledged, with the sender increasing the congestion window by 1 MSS for each of
the acknowledged segments, giving a congestion window of 4 MSS, and so on.
This process results in a doubling of the sending rate every RTT. Thus, the TCP
send rate starts slow but grows exponentially during the slow start phase.
But when should this exponential growth end? Slow start provides several
answers to this question. First, if there is a loss event (i.e., congestion) indicated
by a timeout, the TCP sender sets the value of cwnd to 1 and begins the slow
start process anew. It also sets the value of a second state variable, ssthresh
(shorthand for “slow start threshold”) to cwnd/2—half of the value of the congestion window value when congestion was detected. The second way in which
slow start may end is directly tied to the value of ssthresh. Since ssthresh
is half the value of cwnd when congestion was last detected, it might be a bit
reckless to keep doubling cwnd when it reaches or surpasses the value of
ssthresh. Thus, when the value of cwnd equals ssthresh, slow start ends
and TCP transitions into congestion avoidance mode. As we’ll see, TCP increases
3.7
•
TCP CONGESTION CONTROL
PRINCIPLES IN PRACTICE
TCP SPLITTING: OPTIMIZING THE PERFORMANCE OF CLOUD SERVICES
For cloud services such as search, e-mail, and social networks, it is desirable to provide a
high-level of responsiveness, ideally giving users the illusion that the services are running
within their own end systems (including their smartphones). This can be a major challenge,
as users are often located far away from the data centers that are responsible for serving
the dynamic content associated with the cloud services. Indeed, if the end system is far
from a data center, then the RTT will be large, potentially leading to poor response time
performance due to TCP slow start.
As a case study, consider the delay in receiving a response for a search query.
Typically, the server requires three TCP windows during slow start to deliver the
response [Pathak 2010]. Thus the time from when an end system initiates a TCP connection until the time when it receives the last packet of the response is roughly 4 ⭈ RTT
(one RTT to set up the TCP connection plus three RTTs for the three windows of data)
plus the processing time in the data center. These RTT delays can lead to a noticeable
delay in returning search results for a significant fraction of queries. Moreover, there
can be significant packet loss in access networks, leading to TCP retransmissions and
even larger delays.
One way to mitigate this problem and improve user-perceived performance is to (1)
deploy front-end servers closer to the users, and (2) utilize TCP splitting by breaking the
TCP connection at the front-end server. With TCP splitting, the client establishes a TCP connection to the nearby front-end, and the front-end maintains a persistent TCP connection to
the data center with a very large TCP congestion window [Tariq 2008, Pathak 2010,
Chen 2011]. With this approach, the response time roughly becomes 4 ⭈ RTTFE + RTTBE +
processing time, where RTTFE is the round-trip time between client and front-end server, and
RTTBE is the round-trip time between the front-end server and the data center (back-end server).
If the front-end server is close to client, then this response time approximately becomes
RTT plus processing time, since RTTFE is negligibly small and RTTBE is approximately RTT. In
summary, TCP splitting can reduce the networking delay roughly from 4 ⭈ RTT to RTT, significantly improving user-perceived performance, particularly for users who are far from the
nearest data center. TCP splitting also helps reduce TCP retransmission delays caused by
losses in access networks. Today, Google and Akamai make extensive use of their CDN
servers in access networks (see Section 7.2) to perform TCP splitting for the cloud services
they support [Chen 2011].
cwnd more cautiously when in congestion-avoidance mode. The final way in
which slow start can end is if three duplicate ACKs are detected, in which case
TCP performs a fast retransmit (see Section 3.5.4) and enters the fast recovery
state, as discussed below. TCP’s behavior in slow start is summarized in the FSM
273
274
CHAPTER 3
•
TRANSPORT LAYER
Host A
Host B
one se
gmen
t
RTT
two se
gmen
ts
four s
egment
Time
Figure 3.51
s
Time
TCP slow start
description of TCP congestion control in Figure 3.52. The slow-start algorithm
traces it roots to [Jacobson 1988]; an approach similar to slow start was also proposed independently in [Jain 1986].
Congestion Avoidance
On entry to the congestion-avoidance state, the value of cwnd is approximately half
its value when congestion was last encountered—congestion could be just around
the corner! Thus, rather than doubling the value of cwnd every RTT, TCP adopts a
more conservative approach and increases the value of cwnd by just a single MSS
every RTT [RFC 5681]. This can be accomplished in several ways. A common
approach is for the TCP sender to increase cwnd by MSS bytes (MSS/cwnd) whenever a new acknowledgment arrives. For example, if MSS is 1,460 bytes and cwnd
is 14,600 bytes, then 10 segments are being sent within an RTT. Each arriving ACK
(assuming one ACK per segment) increases the congestion window size by 1/10
3.7
duplicate ACK
dupACKcount++
Λ
•
TCP CONGESTION CONTROL
new ACK
new ACK
cwnd=cwnd+MSS
dupACKcount=0
transmit new segment(s), as allowed
cwnd=cwnd+MSS •(MSS/cwnd)
dupACKcount=0
transmit new segment(s), as allowed
cwnd=1 MSS
ssthresh =64 KB
dupACKcount=0
cwnd ≥ ssthresh
Λ
Slow
start
Congestion
avoidance
timeout
timeout
ssthresh=cwnd/2
cwnd=1 MSS
dupACKcount=0
retransmit missing segment
ssthresh=cwnd/2
cwnd=1 MSS
dupACKcount=0
retransmit missing segment
duplicate ACK
dupACKcount++
timeout
dupACKcount==3
ssthresh=cwnd/2
cwnd=1
dupACKcount=0
retransmit missing segment
new ACK
cwnd=ssthresh
dupACKcount=0
dupACKcount==3
ssthresh=cwnd/2
cwnd=ssthresh+3•MSS
retransmit missing segment
ssthresh=cwnd/2
cwnd=ssthresh+3•MSS
retransmit missing segment
Fast
recovery
duplicate ACK
cwnd=cwnd+MSS
transmit new segment(s), as allowed
Figure 3.52
275
FSM description of TCP congestion control
MSS, and thus, the value of the congestion window will have increased by one MSS
after ACKs when all 10 segments have been received.
But when should congestion avoidance’s linear increase (of 1 MSS per RTT)
end? TCP’s congestion-avoidance algorithm behaves the same when a timeout
occurs. As in the case of slow start: The value of cwnd is set to 1 MSS, and the
value of ssthresh is updated to half the value of cwnd when the loss event
occurred. Recall, however, that a loss event also can be triggered by a triple duplicate ACK event. In this case, the network is continuing to deliver segments from
sender to receiver (as indicated by the receipt of duplicate ACKs). So TCP’s behavior to this type of loss event should be less drastic than with a timeout-indicated loss:
TCP halves the value of cwnd (adding in 3 MSS for good measure to account for
276
CHAPTER 3
•
TRANSPORT LAYER
the triple duplicate ACKs received) and records the value of ssthresh to be half
the value of cwnd when the triple duplicate ACKs were received. The fast-recovery
state is then entered.
Fast Recovery
16
14
Congestion window
(in segments)
VideoNote
Examining the
behavior of TCP
In fast recovery, the value of cwnd is increased by 1 MSS for every duplicate ACK
received for the missing segment that caused TCP to enter the fast-recovery state.
Eventually, when an ACK arrives for the missing segment, TCP enters the
congestion-avoidance state after deflating cwnd. If a timeout event occurs, fast
recovery transitions to the slow-start state after performing the same actions as in
slow start and congestion avoidance: The value of cwnd is set to 1 MSS, and the
value of ssthresh is set to half the value of cwnd when the loss event occurred.
Fast recovery is a recommended, but not required, component of TCP [RFC
5681]. It is interesting that an early version of TCP, known as TCP Tahoe, unconditionally cut its congestion window to 1 MSS and entered the slow-start phase after
either a timeout-indicated or triple-duplicate-ACK-indicated loss event. The newer
version of TCP, TCP Reno, incorporated fast recovery.
Figure 3.53 illustrates the evolution of TCP’s congestion window for both Reno
and Tahoe. In this figure, the threshold is initially equal to 8 MSS. For the first eight
transmission rounds, Tahoe and Reno take identical actions. The congestion window
climbs exponentially fast during slow start and hits the threshold at the fourth round
of transmission. The congestion window then climbs linearly until a triple duplicateACK event occurs, just after transmission round 8. Note that the congestion window
is 12 • MSS when this loss event occurs. The value of ssthresh is then set to
12
TCP Reno
10
ssthresh
8
6
ssthresh
4
TCP Tahoe
2
0
0
Figure 3.53
1
2
3
4
5 6 7 8 9 10 11 12 13 14 15
Transmission round
Evolution of TCP’s congestion window (Tahoe and Reno)
3.7
•
TCP CONGESTION CONTROL
0.5 • cwnd = 6 • MSS. Under TCP Reno, the congestion window is set to cwnd =
6 • MSS and then grows linearly. Under TCP Tahoe, the congestion window is set to
1 MSS and grows exponentially until it reaches the value of ssthresh, at which
point it grows linearly.
Figure 3.52 presents the complete FSM description of TCP’s congestioncontrol algorithms—slow start, congestion avoidance, and fast recovery. The figure
also indicates where transmission of new segments or retransmitted segments can
occur. Although it is important to distinguish between TCP error control/retransmission and TCP congestion control, it’s also important to appreciate how these two
aspects of TCP are inextricably linked.
TCP Congestion Control: Retrospective
Having delved into the details of slow start, congestion avoidance, and fast recovery, it’s worthwhile to now step back and view the forest from the trees. Ignoring the
initial slow-start period when a connection begins and assuming that losses are indicated by triple duplicate ACKs rather than timeouts, TCP’s congestion control consists of linear (additive) increase in cwnd of 1 MSS per RTT and then a halving
(multiplicative decrease) of cwnd on a triple duplicate-ACK event. For this reason,
TCP congestion control is often referred to as an additive-increase, multiplicativedecrease (AIMD) form of congestion control. AIMD congestion control gives rise
to the “saw tooth” behavior shown in Figure 3.54, which also nicely illustrates our
earlier intuition of TCP “probing” for bandwidth—TCP linearly increases its congestion window size (and hence its transmission rate) until a triple duplicate-ACK
event occurs. It then decreases its congestion window size by a factor of two but
then again begins increasing it linearly, probing to see if there is additional available
bandwidth.
Congestion window
24 K
16 K
8K
Time
Figure 3.54
Additive-increase, multiplicative-decrease congestion control
277
278
CHAPTER 3
•
TRANSPORT LAYER
As noted previously, many TCP implementations use the Reno algorithm [Padhye
2001]. Many variations of the Reno algorithm have been proposed [RFC 3782; RFC
2018]. The TCP Vegas algorithm [Brakmo 1995; Ahn 1995] attempts to avoid congestion while maintaining good throughput. The basic idea of Vegas is to (1) detect congestion in the routers between source and destination before packet loss occurs, and (2)
lower the rate linearly when this imminent packet loss is detected. Imminent packet loss
is predicted by observing the RTT. The longer the RTT of the packets, the greater the
congestion in the routers. Linux supports a number of congestion-control algorithms
(including TCP Reno and TCP Vegas) and allows a system administrator to configure
which version of TCP will be used. The default version of TCP in Linux version 2.6.18
was set to CUBIC [Ha 2008], a version of TCP developed for high-bandwidth applications. For a recent survey of the many flavors of TCP, see [Afanasyev 2010].
TCP’s AIMD algorithm was developed based on a tremendous amount of engineering insight and experimentation with congestion control in operational networks. Ten years after TCP’s development, theoretical analyses showed that TCP’s
congestion-control algorithm serves as a distributed asynchronous-optimization
algorithm that results in several important aspects of user and network performance
being simultaneously optimized [Kelly 1998]. A rich theory of congestion control
has since been developed [Srikant 2004].
Macroscopic Description of TCP Throughput
Given the saw-toothed behavior of TCP, it’s natural to consider what the average
throughput (that is, the average rate) of a long-lived TCP connection might be. In this
analysis we’ll ignore the slow-start phases that occur after timeout events. (These
phases are typically very short, since the sender grows out of the phase exponentially
fast.) During a particular round-trip interval, the rate at which TCP sends data is a
function of the congestion window and the current RTT. When the window size is w
bytes and the current round-trip time is RTT seconds, then TCP’s transmission rate is
roughly w/RTT. TCP then probes for additional bandwidth by increasing w by 1 MSS
each RTT until a loss event occurs. Denote by W the value of w when a loss event
occurs. Assuming that RTT and W are approximately constant over the duration of
the connection, the TCP transmission rate ranges from W/(2 · RTT) to W/RTT.
These assumptions lead to a highly simplified macroscopic model for the
steady-state behavior of TCP. The network drops a packet from the connection when
the rate increases to W/RTT; the rate is then cut in half and then increases by
MSS/RTT every RTT until it again reaches W/RTT. This process repeats itself over
and over again. Because TCP’s throughput (that is, rate) increases linearly between
the two extreme values, we have
average throughput of a connection =
0.75 ⭈ W
RTT
3.7
•
TCP CONGESTION CONTROL
Using this highly idealized model for the steady-state dynamics of TCP, we can
also derive an interesting expression that relates a connection’s loss rate to its available bandwidth [Mahdavi 1997]. This derivation is outlined in the homework problems. A more sophisticated model that has been found empirically to agree with
measured data is [Padhye 2000].
TCP Over High-Bandwidth Paths
It is important to realize that TCP congestion control has evolved over the years and
indeed continues to evolve. For a summary of current TCP variants and discussion
of TCP evolution, see [Floyd 2001, RFC 5681, Afanasyev 2010]. What was good
for the Internet when the bulk of the TCP connections carried SMTP, FTP, and Telnet traffic is not necessarily good for today’s HTTP-dominated Internet or for a
future Internet with services that are still undreamed of.
The need for continued evolution of TCP can be illustrated by considering the
high-speed TCP connections that are needed for grid- and cloud-computing applications. For example, consider a TCP connection with 1,500-byte segments and a 100
ms RTT, and suppose we want to send data through this connection at 10 Gbps.
Following [RFC 3649], we note that using the TCP throughput formula above, in
order to achieve a 10 Gbps throughput, the average congestion window size would
need to be 83,333 segments. That’s a lot of segments, leading us to be rather concerned that one of these 83,333 in-flight segments might be lost. What would happen
in the case of a loss? Or, put another way, what fraction of the transmitted segments
could be lost that would allow the TCP congestion-control algorithm specified in Figure 3.52 still to achieve the desired 10 Gbps rate? In the homework questions for this
chapter, you are led through the derivation of a formula relating the throughput of a
TCP connection as a function of the loss rate (L), the round-trip time (RTT), and the
maximum segment size (MSS):
average throughput of a connection =
1.22 ⭈ MSS
RTT 2L
Using this formula, we can see that in order to achieve a throughput of 10 Gbps,
today’s TCP congestion-control algorithm can only tolerate a segment loss probability of 2 · 10–10 (or equivalently, one loss event for every 5,000,000,000 segments)—
a very low rate. This observation has led a number of researchers to investigate new
versions of TCP that are specifically designed for such high-speed environments;
see [Jin 2004; RFC 3649; Kelly 2003; Ha 2008] for discussions of these efforts.
3.7.1 Fairness
Consider K TCP connections, each with a different end-to-end path, but all passing
through a bottleneck link with transmission rate R bps. (By bottleneck link, we mean
279
280
CHAPTER 3
•
TRANSPORT LAYER
that for each connection, all the other links along the connection’s path are not congested and have abundant transmission capacity as compared with the transmission
capacity of the bottleneck link.) Suppose each connection is transferring a large file
and there is no UDP traffic passing through the bottleneck link. A congestion-control mechanism is said to be fair if the average transmission rate of each connection
is approximately R/K; that is, each connection gets an equal share of the link bandwidth.
Is TCP’s AIMD algorithm fair, particularly given that different TCP connections may start at different times and thus may have different window sizes at a
given point in time? [Chiu 1989] provides an elegant and intuitive explanation of
why TCP congestion control converges to provide an equal share of a bottleneck
link’s bandwidth among competing TCP connections.
Let’s consider the simple case of two TCP connections sharing a single link
with transmission rate R, as shown in Figure 3.55. Assume that the two connections
have the same MSS and RTT (so that if they have the same congestion window size,
then they have the same throughput), that they have a large amount of data to send,
and that no other TCP connections or UDP datagrams traverse this shared link. Also,
ignore the slow-start phase of TCP and assume the TCP connections are operating
in CA mode (AIMD) at all times.
Figure 3.56 plots the throughput realized by the two TCP connections. If TCP is
to share the link bandwidth equally between the two connections, then the realized
throughput should fall along the 45-degree arrow (equal bandwidth share) emanating from the origin. Ideally, the sum of the two throughputs should equal R. (Certainly, each connection receiving an equal, but zero, share of the link capacity is not
a desirable situation!) So the goal should be to have the achieved throughputs fall
somewhere near the intersection of the equal bandwidth share line and the full bandwidth utilization line in Figure 3.56.
Suppose that the TCP window sizes are such that at a given point in time, connections 1 and 2 realize throughputs indicated by point A in Figure 3.56. Because
the amount of link bandwidth jointly consumed by the two connections is less than
TCP connection 2
Bottleneck
router capacity R
TCP connection 1
Figure 3.55
Two TCP connections sharing a single bottleneck link
3.7
•
TCP CONGESTION CONTROL
R
Connection 2 throughput
Full bandwidth
utilization line
Equal
bandwidth
share
D
B
C
A
Connection 1 throughput
Figure 3.56
R
Throughput realized by TCP connections 1 and 2
R, no loss will occur, and both connections will increase their window by 1 MSS
per RTT as a result of TCP’s congestion-avoidance algorithm. Thus, the joint
throughput of the two connections proceeds along a 45-degree line (equal increase
for both connections) starting from point A. Eventually, the link bandwidth jointly
consumed by the two connections will be greater than R, and eventually packet loss
will occur. Suppose that connections 1 and 2 experience packet loss when they
realize throughputs indicated by point B. Connections 1 and 2 then decrease their
windows by a factor of two. The resulting throughputs realized are thus at point C,
halfway along a vector starting at B and ending at the origin. Because the joint
bandwidth use is less than R at point C, the two connections again increase their
throughputs along a 45-degree line starting from C. Eventually, loss will again
occur, for example, at point D, and the two connections again decrease their window sizes by a factor of two, and so on. You should convince yourself that the
bandwidth realized by the two connections eventually fluctuates along the equal
bandwidth share line. You should also convince yourself that the two connections
will converge to this behavior regardless of where they are in the two-dimensional
space! Although a number of idealized assumptions lie behind this scenario, it still
provides an intuitive feel for why TCP results in an equal sharing of bandwidth
among connections.
In our idealized scenario, we assumed that only TCP connections traverse
the bottleneck link, that the connections have the same RTT value, and that only a
281
282
CHAPTER 3
•
TRANSPORT LAYER
single TCP connection is associated with a host-destination pair. In practice, these
conditions are typically not met, and client-server applications can thus obtain very
unequal portions of link bandwidth. In particular, it has been shown that when multiple connections share a common bottleneck, those sessions with a smaller RTT are
able to grab the available bandwidth at that link more quickly as it becomes free
(that is, open their congestion windows faster) and thus will enjoy higher throughput than those connections with larger RTTs [Lakshman 1997].
Fairness and UDP
We have just seen how TCP congestion control regulates an application’s transmission rate via the congestion window mechanism. Many multimedia applications,
such as Internet phone and video conferencing, often do not run over TCP for this
very reason—they do not want their transmission rate throttled, even if the network
is very congested. Instead, these applications prefer to run over UDP, which does
not have built-in congestion control. When running over UDP, applications can
pump their audio and video into the network at a constant rate and occasionally lose
packets, rather than reduce their rates to “fair” levels at times of congestion and not
lose any packets. From the perspective of TCP, the multimedia applications running
over UDP are not being fair—they do not cooperate with the other connections nor
adjust their transmission rates appropriately. Because TCP congestion control will
decrease its transmission rate in the face of increasing congestion (loss), while UDP
sources need not, it is possible for UDP sources to crowd out TCP traffic. An area of
research today is thus the development of congestion-control mechanisms for the
Internet that prevent UDP traffic from bringing the Internet’s throughput to a grinding halt [Floyd 1999; Floyd 2000; Kohler 2006].
Fairness and Parallel TCP Connections
But even if we could force UDP traffic to behave fairly, the fairness problem would
still not be completely solved. This is because there is nothing to stop a TCP-based
application from using multiple parallel connections. For example, Web browsers
often use multiple parallel TCP connections to transfer the multiple objects within
a Web page. (The exact number of multiple connections is configurable in most
browsers.) When an application uses multiple parallel connections, it gets a larger
fraction of the bandwidth in a congested link. As an example, consider a link of rate
R supporting nine ongoing client-server applications, with each of the applications
using one TCP connection. If a new application comes along and also uses one
TCP connection, then each application gets approximately the same transmission
rate of R/10. But if this new application instead uses 11 parallel TCP connections,
then the new application gets an unfair allocation of more than R/2. Because
Web traffic is so pervasive in the Internet, multiple parallel connections are not
uncommon.
3.8
•
3.8 Summary
We began this chapter by studying the services that a transport-layer protocol can
provide to network applications. At one extreme, the transport-layer protocol can be
very simple and offer a no-frills service to applications, providing only a multiplexing/demultiplexing function for communicating processes. The Internet’s UDP protocol is an example of such a no-frills transport-layer protocol. At the other extreme,
a transport-layer protocol can provide a variety of guarantees to applications, such
as reliable delivery of data, delay guarantees, and bandwidth guarantees. Nevertheless, the services that a transport protocol can provide are often constrained by the
service model of the underlying network-layer protocol. If the network-layer protocol cannot provide delay or bandwidth guarantees to transport-layer segments, then
the transport-layer protocol cannot provide delay or bandwidth guarantees for the
messages sent between processes.
We learned in Section 3.4 that a transport-layer protocol can provide reliable
data transfer even if the underlying network layer is unreliable. We saw that providing reliable data transfer has many subtle points, but that the task can be accomplished by carefully combining acknowledgments, timers, retransmissions, and
sequence numbers.
Although we covered reliable data transfer in this chapter, we should keep in
mind that reliable data transfer can be provided by link-, network-, transport-, or
application-layer protocols. Any of the upper four layers of the protocol stack can
implement acknowledgments, timers, retransmissions, and sequence numbers and
provide reliable data transfer to the layer above. In fact, over the years, engineers
and computer scientists have independently designed and implemented link-, network-, transport-, and application-layer protocols that provide reliable data transfer
(although many of these protocols have quietly disappeared).
In Section 3.5, we took a close look at TCP, the Internet’s connection-oriented
and reliable transport-layer protocol. We learned that TCP is complex, involving
connection management, flow control, and round-trip time estimation, as well as
reliable data transfer. In fact, TCP is actually more complex than our description—
we intentionally did not discuss a variety of TCP patches, fixes, and improvements
that are widely implemented in various versions of TCP. All of this complexity,
however, is hidden from the network application. If a client on one host wants to
send data reliably to a server on another host, it simply opens a TCP socket to the
server and pumps data into that socket. The client-server application is blissfully
unaware of TCP’s complexity.
In Section 3.6, we examined congestion control from a broad perspective, and in
Section 3.7, we showed how TCP implements congestion control. We learned that
congestion control is imperative for the well-being of the network. Without congestion control, a network can easily become gridlocked, with little or no data being transported end-to-end. In Section 3.7 we learned that TCP implements an end-to-end
SUMMARY
283
284
CHAPTER 3
•
TRANSPORT LAYER
congestion-control mechanism that additively increases its transmission rate when the
TCP connection’s path is judged to be congestion-free, and multiplicatively decreases
its transmission rate when loss occurs. This mechanism also strives to give each TCP
connection passing through a congested link an equal share of the link bandwidth. We
also examined in some depth the impact of TCP connection establishment and slow
start on latency. We observed that in many important scenarios, connection establishment and slow start significantly contribute to end-to-end delay. We emphasize once
more that while TCP congestion control has evolved over the years, it remains an area
of intensive research and will likely continue to evolve in the upcoming years.
Our discussion of specific Internet transport protocols in this chapter has
focused on UDP and TCP—the two “work horses” of the Internet transport layer.
However, two decades of experience with these two protocols has identified
circumstances in which neither is ideally suited. Researchers have thus been
busy developing additional transport-layer protocols, several of which are now
IETF proposed standards.
The Datagram Congestion Control Protocol (DCCP) [RFC 4340] provides a lowoverhead, message-oriented, UDP-like unreliable service, but with an applicationselected form of congestion control that is compatible with TCP. If reliable or
semi-reliable data transfer is needed by an application, then this would be performed
within the application itself, perhaps using the mechanisms we have studied in Section
3.4. DCCP is envisioned for use in applications such as streaming media (see Chapter 7)
that can exploit the tradeoff between timeliness and reliability of data delivery, but that
want to be responsive to network congestion.
The Stream Control Transmission Protocol (SCTP) [RFC 4960, RFC 3286] is a
reliable, message-oriented protocol that allows several different application-level
“streams” to be multiplexed through a single SCTP connection (an approach known as
“multi-streaming”). From a reliability standpoint, the different streams within the connection are handled separately, so that packet loss in one stream does not affect the
delivery of data in other streams. SCTP also allows data to be transferred over two outgoing paths when a host is connected to two or more networks, optional delivery of outof-order data, and a number of other features. SCTP’s flow- and congestion-control
algorithms are essentially the same as in TCP.
The TCP-Friendly Rate Control (TFRC) protocol [RFC 5348] is a congestioncontrol protocol rather than a full-fledged transport-layer protocol. It specifies a
congestion-control mechanism that could be used in anther transport protocol such as
DCCP (indeed one of the two application-selectable protocols available in DCCP is
TFRC). The goal of TFRC is to smooth out the “saw tooth” behavior (see Figure 3.54)
in TCP congestion control, while maintaining a long-term sending rate that is “reasonably” close to that of TCP. With a smoother sending rate than TCP, TFRC is well-suited
for multimedia applications such as IP telephony or streaming media where such a
smooth rate is important. TFRC is an “equation-based” protocol that uses the measured
packet loss rate as input to an equation [Padhye 2000] that estimates what TCP’s
throughput would be if a TCP session experiences that loss rate. This rate is then taken
as TFRC’s target sending rate.
HOMEWORK PROBLEMS AND QUESTIONS
Only the future will tell whether DCCP, SCTP, or TFRC will see widespread
deployment. While these protocols clearly provide enhanced capabilities over TCP and
UDP, TCP and UDP have proven themselves “good enough” over the years. Whether
“better” wins out over “good enough” will depend on a complex mix of technical,
social, and business considerations.
In Chapter 1, we said that a computer network can be partitioned into the
“network edge” and the “network core.” The network edge covers everything that
happens in the end systems. Having now covered the application layer and the
transport layer, our discussion of the network edge is complete. It is time to
explore the network core! This journey begins in the next chapter, where we’ll
study the network layer, and continues into Chapter 5, where we’ll study the
link layer.
Homework Problems and Questions
Chapter 3 Review Questions
SECTIONS 3.1–3.3
R1. Suppose the network layer provides the following service. The network layer
in the source host accepts a segment of maximum size 1,200 bytes and a destination host address from the transport layer. The network layer then guarantees to deliver the segment to the transport layer at the destination host.
Suppose many network application processes can be running at the
destination host.
a. Design the simplest possible transport-layer protocol that will get application data to the desired process at the destination host. Assume the operating system in the destination host has assigned a 4-byte port number to
each running application process.
b. Modify this protocol so that it provides a “return address” to the destination process.
c. In your protocols, does the transport layer “have to do anything” in the
core of the computer network?
R2. Consider a planet where everyone belongs to a family of six, every family
lives in its own house, each house has a unique address, and each person in a
given house has a unique name. Suppose this planet has a mail service that
delivers letters from source house to destination house. The mail service
requires that (1) the letter be in an envelope, and that (2) the address of the
destination house (and nothing more) be clearly written on the envelope. Suppose each family has a delegate family member who collects and distributes
letters for the other family members. The letters do not necessarily provide
any indication of the recipients of the letters.
285
286
CHAPTER 3
•
TRANSPORT LAYER
R3.
R4.
R5.
R6.
R7.
R8.
a. Using the solution to Problem R1 above as inspiration, describe a protocol
that the delegates can use to deliver letters from a sending family member
to a receiving family member.
b. In your protocol, does the mail service ever have to open the envelope and
examine the letter in order to provide its service?
Consider a TCP connection between Host A and Host B. Suppose that the
TCP segments traveling from Host A to Host B have source port number x
and destination port number y. What are the source and destination port numbers for the segments traveling from Host B to Host A?
Describe why an application developer might choose to run an application
over UDP rather than TCP.
Why is it that voice and video traffic is often sent over TCP rather than UDP
in today’s Internet? (Hint: The answer we are looking for has nothing to do
with TCP’s congestion-control mechanism.)
Is it possible for an application to enjoy reliable data transfer even when the
application runs over UDP? If so, how?
Suppose a process in Host C has a UDP socket with port number 6789. Suppose both Host A and Host B each send a UDP segment to Host C with destination port number 6789. Will both of these segments be directed to the same
socket at Host C? If so, how will the process at Host C know that these two
segments originated from two different hosts?
Suppose that a Web server runs in Host C on port 80. Suppose this Web
server uses persistent connections, and is currently receiving requests from
two different Hosts, A and B. Are all of the requests being sent through the
same socket at Host C? If they are being passed through different sockets, do
both of the sockets have port 80? Discuss and explain.
SECTION 3.4
R9. In our rdt protocols, why did we need to introduce sequence numbers?
R10. In our rdt protocols, why did we need to introduce timers?
R11. Suppose that the roundtrip delay between sender and receiver is constant and
known to the sender. Would a timer still be necessary in protocol rdt 3.0,
assuming that packets can be lost? Explain.
R12. Visit the Go-Back-N Java applet at the companion Web site.
a. Have the source send five packets, and then pause the animation before
any of the five packets reach the destination. Then kill the first packet and
resume the animation. Describe what happens.
b. Repeat the experiment, but now let the first packet reach the destination
and kill the first acknowledgment. Describe again what happens.
c. Finally, try sending six packets. What happens?
HOMEWORK PROBLEMS AND QUESTIONS
R13. Repeat R12, but now with the Selective Repeat Java applet. How are Selective Repeat and Go-Back-N different?
SECTION 3.5
R14. True or false?
a. Host A is sending Host B a large file over a TCP connection. Assume
Host B has no data to send Host A. Host B will not send acknowledgments to Host A because Host B cannot piggyback the acknowledgments
on data.
b. The size of the TCP rwnd never changes throughout the duration of the
connection.
c. Suppose Host A is sending Host B a large file over a TCP connection. The
number of unacknowledged bytes that A sends cannot exceed the size of
the receive buffer.
d. Suppose Host A is sending a large file to Host B over a TCP connection. If
the sequence number for a segment of this connection is m, then the
sequence number for the subsequent segment will necessarily be m + 1.
e. The TCP segment has a field in its header for rwnd.
f. Suppose that the last SampleRTT in a TCP connection is equal to 1 sec.
The current value of TimeoutInterval for the connection will necessarily be ≥ 1 sec.
g. Suppose Host A sends one segment with sequence number 38 and 4 bytes
of data over a TCP connection to Host B. In this same segment the
acknowledgment number is necessarily 42.
R15. Suppose Host A sends two TCP segments back to back to Host B over a TCP
connection. The first segment has sequence number 90; the second has
sequence number 110.
a. How much data is in the first segment?
b. Suppose that the first segment is lost but the second segment arrives at B.
In the acknowledgment that Host B sends to Host A, what will be the
acknowledgment number?
R16. Consider the Telnet example discussed in Section 3.5. A few seconds after the
user types the letter ‘C,’ the user types the letter ‘R.’ After typing the letter
‘R,’ how many segments are sent, and what is put in the sequence number
and acknowledgment fields of the segments?
SECTION 3.7
R17. Suppose two TCP connections are present over some bottleneck link of rate R
bps. Both connections have a huge file to send (in the same direction over the
287
288
CHAPTER 3
•
TRANSPORT LAYER
bottleneck link). The transmissions of the files start at the same time. What
transmission rate would TCP like to give to each of the connections?
R18. True or false? Consider congestion control in TCP. When the timer expires at
the sender, the value of ssthresh is set to one half of its previous value.
R19. In the discussion of TCP splitting in the sidebar in Section 7.2, it was
claimed that the response time with TCP splitting is approximately
4 ⭈ RTTFE + RTTBE + processing time. Justify this claim.
Problems
P1. Suppose Client A initiates a Telnet session with Server S. At about the same
time, Client B also initiates a Telnet session with Server S. Provide possible
source and destination port numbers for
a. The segments sent from A to S.
b. The segments sent from B to S.
c. The segments sent from S to A.
d. The segments sent from S to B.
e. If A and B are different hosts, is it possible that the source port number in
the segments from A to S is the same as that from B to S?
f. How about if they are the same host?
P2. Consider Figure 3.5. What are the source and destination port values in the segments flowing from the server back to the clients’ processes? What are the IP
addresses in the network-layer datagrams carrying the transport-layer segments?
P3. UDP and TCP use 1s complement for their checksums. Suppose you have the
following three 8-bit bytes: 01010011, 01100110, 01110100. What is the 1s
complement of the sum of these 8-bit bytes? (Note that although UDP and
TCP use 16-bit words in computing the checksum, for this problem you are
being asked to consider 8-bit sums.) Show all work. Why is it that UDP takes
the 1s complement of the sum; that is, why not just use the sum? With the 1s
complement scheme, how does the receiver detect errors? Is it possible that a
1-bit error will go undetected? How about a 2-bit error?
P4. a. Suppose you have the following 2 bytes: 01011100 and 01100101. What is
the 1s complement of the sum of these 2 bytes?
b. Suppose you have the following 2 bytes: 11011010 and 01100101. What is
the 1s complement of the sum of these 2 bytes?
c. For the bytes in part (a), give an example where one bit is flipped in each
of the 2 bytes and yet the 1s complement doesn’t change.
PROBLEMS
P5. Suppose that the UDP receiver computes the Internet checksum for the received
UDP segment and finds that it matches the value carried in the checksum field.
Can the receiver be absolutely certain that no bit errors have occurred? Explain.
P6. Consider our motivation for correcting protocol rdt2.1. Show that the
receiver, shown in Figure 3.57, when operating with the sender shown in Figure 3.11, can lead the sender and receiver to enter into a deadlock state, where
each is waiting for an event that will never occur.
P7. In protocol rdt3.0, the ACK packets flowing from the receiver to the
sender do not have sequence numbers (although they do have an ACK field
that contains the sequence number of the packet they are acknowledging).
Why is it that our ACK packets do not require sequence numbers?
P8. Draw the FSM for the receiver side of protocol rdt3.0.
P9. Give a trace of the operation of protocol rdt3.0 when data packets and
acknowledgment packets are garbled. Your trace should be similar to that
used in Figure 3.16.
P10. Consider a channel that can lose packets but has a maximum delay that is
known. Modify protocol rdt2.1 to include sender timeout and retransmit.
Informally argue why your protocol can communicate correctly over this
channel.
rdt_rvc(rcvpkt) && notcorrupt(rcvpkt)
&& has_seq0(rcvpkt)
extract(rcvpkt,data)
deliver_data(data)
compute chksum
make_pkt(sendpkt,ACK,chksum)
udt_send(sndpkt)
rdt_rcv(rcvpkt) &&
(corrupt(rcvpkt)||
has_seq1(rcvpkt)))
compute chksum
make_pkt(sndpkt,NAK,chksum)
udt_send(sndpkt)
Wait for
0 from
below
Wait for
1 from
below
rdt_rvc(rcvpkt) && notcorrupt(rcvpkt)
&& has_seq1(rcvpkt)
extract(rcvpkt,data)
deliver_data(data)
compute chksum
make_pkt(sendpkt,ACK,chksum)
udt_send(sndpkt)
Figure 3.57
An incorrect receiver for protocol rdt 2.1
rdt_rcv(rcvpkt) &&
(corrupt(rcvpkt)||
has_seq0(rcvpkt)))
compute chksum
make_pkt(sndpkt,NAK,chksum)
udt_send(sndpkt)
289
290
CHAPTER 3
•
TRANSPORT LAYER
P11. Consider the rdt2.2 receiver in Figure 3.14, and the creation of a new packet
in the self-transition (i.e., the transition from the state back to itself) in the Waitfor-0-from-below and the Wait-for-1-from-below states: sndpkt=make_
pkt(ACK,0,checksum) and sndpkt=make_pkt(ACK,0,
checksum). Would the protocol work correctly if this action were removed
from the self-transition in the Wait-for-1-from-below state? Justify your
answer. What if this event were removed from the self-transition in the Waitfor-0-from-below state? [Hint: In this latter case, consider what would happen if the first sender-to-receiver packet were corrupted.]
P12. The sender side of rdt3.0 simply ignores (that is, takes no action on)
all received packets that are either in error or have the wrong value in the
acknum field of an acknowledgment packet. Suppose that in such circumstances, rdt3.0 were simply to retransmit the current data packet. Would
the protocol still work? (Hint: Consider what would happen if there were
only bit errors; there are no packet losses but premature timeouts can occur.
Consider how many times the nth packet is sent, in the limit as n
approaches infinity.)
P13. Consider the rdt 3.0 protocol. Draw a diagram showing that if the network connection between the sender and receiver can reorder messages
(that is, that two messages propagating in the medium between the sender
and receiver can be reordered), then the alternating-bit protocol will not
work correctly (make sure you clearly identify the sense in which it will
not work correctly). Your diagram should have the sender on the left and
the receiver on the right, with the time axis running down the page, showing data (D) and acknowledgment (A) message exchange. Make sure you
indicate the sequence number associated with any data or acknowledgment
segment.
P14. Consider a reliable data transfer protocol that uses only negative acknowledgments. Suppose the sender sends data only infrequently. Would a NAK-only
protocol be preferable to a protocol that uses ACKs? Why? Now suppose the
sender has a lot of data to send and the end-to-end connection experiences
few losses. In this second case, would a NAK-only protocol be preferable to a
protocol that uses ACKs? Why?
P15. Consider the cross-country example shown in Figure 3.17. How big would
the window size have to be for the channel utilization to be greater than 98
percent? Suppose that the size of a packet is 1,500 bytes, including both
header fields and data.
P16. Suppose an application uses rdt 3.0 as its transport layer protocol. As the
stop-and-wait protocol has very low channel utilization (shown in the crosscountry example), the designers of this application let the receiver keep sending back a number (more than two) of alternating ACK 0 and ACK 1 even if
PROBLEMS
the corresponding data have not arrived at the receiver. Would this application design increase the channel utilization? Why? Are there any potential
problems with this approach? Explain.
P17. Consider two network entities, A and B, which are connected by a perfect bidirectional channel (i.e., any message sent will be received correctly; the
channel will not corrupt, lose, or re-order packets). A and B are to deliver
data messages to each other in an alternating manner: First, A must deliver a
message to B, then B must deliver a message to A, then A must deliver a message to B and so on. If an entity is in a state where it should not attempt to
deliver a message to the other side, and there is an event like
rdt_send(data) call from above that attempts to pass data down for
transmission to the other side, this call from above can simply be ignored
with a call to rdt_unable_to_send(data), which informs the higher
layer that it is currently not able to send data. [Note: This simplifying
assumption is made so you don’t have to worry about buffering data.]
Draw a FSM specification for this protocol (one FSM for A, and one FSM for
B!). Note that you do not have to worry about a reliability mechanism here;
the main point of this question is to create a FSM specification that reflects
the synchronized behavior of the two entities. You should use the following
events and actions that have the same meaning as protocol rdt1.0 in
Figure 3.9: rdt_send(data), packet = make_pkt(data),
udt_send(packet), rdt_rcv(packet), extract
(packet,data), deliver_data(data). Make sure your protocol
reflects the strict alternation of sending between A and B. Also, make sure to
indicate the initial states for A and B in your FSM descriptions.
P18. In the generic SR protocol that we studied in Section 3.4.4, the sender transmits a message as soon as it is available (if it is in the window) without waiting for an acknowledgment. Suppose now that we want an SR protocol that
sends messages two at a time. That is, the sender will send a pair of messages
and will send the next pair of messages only when it knows that both messages in the first pair have been received correctly.
Suppose that the channel may lose messages but will not corrupt or reorder
messages. Design an error-control protocol for the unidirectional reliable
transfer of messages. Give an FSM description of the sender and receiver.
Describe the format of the packets sent between sender and receiver, and vice
versa. If you use any procedure calls other than those in Section 3.4 (for
example, udt_send(), start_timer(), rdt_rcv(), and so on),
clearly state their actions. Give an example (a timeline trace of sender and
receiver) showing how your protocol recovers from a lost packet.
P19. Consider a scenario in which Host A wants to simultaneously send packets to
Hosts B and C. A is connected to B and C via a broadcast channel—a packet
291
292
CHAPTER 3
•
TRANSPORT LAYER
sent by A is carried by the channel to both B and C. Suppose that the broadcast channel connecting A, B, and C can independently lose and corrupt
packets (and so, for example, a packet sent from A might be correctly
received by B, but not by C). Design a stop-and-wait-like error-control protocol for reliably transferring packets from A to B and C, such that A will not
get new data from the upper layer until it knows that both B and C have correctly received the current packet. Give FSM descriptions of A and C. (Hint:
The FSM for B should be essentially the same as for C.) Also, give a description of the packet format(s) used.
P20. Consider a scenario in which Host A and Host B want to send messages to
Host C. Hosts A and C are connected by a channel that can lose and corrupt
(but not reorder) messages. Hosts B and C are connected by another channel (independent of the channel connecting A and C) with the same properties. The transport layer at Host C should alternate in delivering messages
from A and B to the layer above (that is, it should first deliver the data from
a packet from A, then the data from a packet from B, and so on). Design a
stop-and-wait-like error-control protocol for reliably transferring packets
from A and B to C, with alternating delivery at C as described above. Give
FSM descriptions of A and C. (Hint: The FSM for B should be essentially
the same as for A.) Also, give a description of the packet format(s) used.
P21. Suppose we have two network entities, A and B. B has a supply of data messages that will be sent to A according to the following conventions. When A
gets a request from the layer above to get the next data (D) message from B,
A must send a request (R) message to B on the A-to-B channel. Only when B
receives an R message can it send a data (D) message back to A on the B-toA channel. A should deliver exactly one copy of each D message to the layer
above. R messages can be lost (but not corrupted) in the A-to-B channel; D
messages, once sent, are always delivered correctly. The delay along both
channels is unknown and variable.
Design (give an FSM description of) a protocol that incorporates the appropriate mechanisms to compensate for the loss-prone A-to-B channel and
implements message passing to the layer above at entity A, as discussed
above. Use only those mechanisms that are absolutely necessary.
P22. Consider the GBN protocol with a sender window size of 4 and a sequence
number range of 1,024. Suppose that at time t, the next in-order packet that the
receiver is expecting has a sequence number of k. Assume that the medium
does not reorder messages. Answer the following questions:
a. What are the possible sets of sequence numbers inside the sender’s window at time t? Justify your answer.
b. What are all possible values of the ACK field in all possible messages currently propagating back to the sender at time t? Justify your answer.
PROBLEMS
P23. Consider the GBN and SR protocols. Suppose the sequence number space is
of size k. What is the largest allowable sender window that will avoid the
occurrence of problems such as that in Figure 3.27 for each of these protocols?
P24. Answer true or false to the following questions and briefly justify your
answer:
a. With the SR protocol, it is possible for the sender to receive an ACK for a
packet that falls outside of its current window.
b. With GBN, it is possible for the sender to receive an ACK for a packet that
falls outside of its current window.
c. The alternating-bit protocol is the same as the SR protocol with a sender
and receiver window size of 1.
d. The alternating-bit protocol is the same as the GBN protocol with a sender
and receiver window size of 1.
P25. We have said that an application may choose UDP for a transport protocol
because UDP offers finer application control (than TCP) of what data is sent
in a segment and when.
a. Why does an application have more control of what data is sent in a segment?
b. Why does an application have more control on when the segment is sent?
P26. Consider transferring an enormous file of L bytes from Host A to Host B.
Assume an MSS of 536 bytes.
a. What is the maximum value of L such that TCP sequence numbers are not
exhausted? Recall that the TCP sequence number field has 4 bytes.
b. For the L you obtain in (a), find how long it takes to transmit the file.
Assume that a total of 66 bytes of transport, network, and data-link header
are added to each segment before the resulting packet is sent out over a
155 Mbps link. Ignore flow control and congestion control so A can pump
out the segments back to back and continuously.
P27. Host A and B are communicating over a TCP connection, and Host B has
already received from A all bytes up through byte 126. Suppose Host A then
sends two segments to Host B back-to-back. The first and second segments
contain 80 and 40 bytes of data, respectively. In the first segment, the
sequence number is 127, the source port number is 302, and the destination
port number is 80. Host B sends an acknowledgment whenever it receives a
segment from Host A.
a. In the second segment sent from Host A to B, what are the sequence number, source port number, and destination port number?
b. If the first segment arrives before the second segment, in the acknowledgment of the first arriving segment, what is the acknowledgment number,
the source port number, and the destination port number?
293
294
CHAPTER 3
•
TRANSPORT LAYER
c. If the second segment arrives before the first segment, in the acknowledgment of the first arriving segment, what is the acknowledgment
number?
d. Suppose the two segments sent by A arrive in order at B. The first acknowledgment is lost and the second acknowledgment arrives after the first timeout interval. Draw a timing diagram, showing these segments and all other
segments and acknowledgments sent. (Assume there is no additional packet
loss.) For each segment in your figure, provide the sequence number and
the number of bytes of data; for each acknowledgment that you add, provide the acknowledgment number.
P28. Host A and B are directly connected with a 100 Mbps link. There is one TCP
connection between the two hosts, and Host A is sending to Host B an enormous file over this connection. Host A can send its application data into its TCP
socket at a rate as high as 120 Mbps but Host B can read out of its TCP receive
buffer at a maximum rate of 50 Mbps. Describe the effect of TCP flow control.
P29. SYN cookies were discussed in Section 3.5.6.
a. Why is it necessary for the server to use a special initial sequence number
in the SYNACK?
b. Suppose an attacker knows that a target host uses SYN cookies. Can the
attacker create half-open or fully open connections by simply sending an
ACK packet to the target? Why or why not?
c. Suppose an attacker collects a large amount of initial sequence numbers sent
by the server. Can the attacker cause the server to create many fully open
connections by sending ACKs with those initial sequence numbers? Why?
P30. Consider the network shown in Scenario 2 in Section 3.6.1. Suppose both
sending hosts A and B have some fixed timeout values.
a. Argue that increasing the size of the finite buffer of the router might possibly decrease the throughput (␭out).
b. Now suppose both hosts dynamically adjust their timeout values (like
what TCP does) based on the buffering delay at the router. Would increasing the buffer size help to increase the throughput? Why?
P31. Suppose that the five measured SampleRTT values (see Section 3.5.3) are
106 ms, 120 ms, 140 ms, 90 ms, and 115 ms. Compute the EstimatedRTT
after each of these SampleRTT values is obtained, using a value of α = 0.125
and assuming that the value of EstimatedRTT was 100 ms just before the
first of these five samples were obtained. Compute also the DevRTT after
each sample is obtained, assuming a value of β = 0.25 and assuming the
value of DevRTT was 5 ms just before the first of these five samples was
obtained. Last, compute the TCP TimeoutInterval after each of these
samples is obtained.
PROBLEMS
P32. Consider the TCP procedure for estimating RTT. Suppose that ␣ = 0.1. Let
SampleRTT1 be the most recent sample RTT, let SampleRTT2 be the next
most recent sample RTT, and so on.
a. For a given TCP connection, suppose four acknowledgments have been
returned with corresponding sample RTTs: SampleRTT4, SampleRTT3,
SampleRTT2, and SampleRTT1. Express EstimatedRTT in terms of
the four sample RTTs.
b. Generalize your formula for n sample RTTs.
c. For the formula in part (b) let n approach infinity. Comment on why this
averaging procedure is called an exponential moving average.
P33. In Section 3.5.3, we discussed TCP’s estimation of RTT. Why do you think
TCP avoids measuring the SampleRTT for retransmitted segments?
P34. What is the relationship between the variable SendBase in Section 3.5.4
and the variable LastByteRcvd in Section 3.5.5?
P35. What is the relationship between the variable LastByteRcvd in Section
3.5.5 and the variable y in Section 3.5.4?
P36. In Section 3.5.4, we saw that TCP waits until it has received three
duplicate ACKs before performing a fast retransmit. Why do you think the
TCP designers chose not to perform a fast retransmit after the first duplicate
ACK for a segment is received?
P37. Compare GBN, SR, and TCP (no delayed ACK). Assume that the timeout
values for all three protocols are sufficiently long such that 5 consecutive data
segments and their corresponding ACKs can be received (if not lost in the
channel) by the receiving host (Host B) and the sending host (Host A) respectively. Suppose Host A sends 5 data segments to Host B, and the 2nd segment
(sent from A) is lost. In the end, all 5 data segments have been correctly
received by Host B.
a. How many segments has Host A sent in total and how many ACKs has
Host B sent in total? What are their sequence numbers? Answer this question for all three protocols.
b. If the timeout values for all three protocol are much longer than 5 RTT,
then which protocol successfully delivers all five data segments in shortest
time interval?
P38. In our description of TCP in Figure 3.53, the value of the threshold,
ssthresh, is set as ssthresh=cwnd/2 in several places and
ssthresh value is referred to as being set to half the window size when a
loss event occurred. Must the rate at which the sender is sending when the
loss event occurred be approximately equal to cwnd segments per RTT?
Explain your answer. If your answer is no, can you suggest a different
manner in which ssthresh should be set?
295
CHAPTER 3
VideoNote
Examining the
behavior of TCP
•
TRANSPORT LAYER
P39. Consider Figure 3.46(b). If ␭⬘in increases beyond R/2, can ␭out increase
beyond R/3? Explain. Now consider Figure 3.46(c). If ␭⬘in increases
beyond R/2, can ␭out increase beyond R/4 under the assumption that a
packet will be forwarded twice on average from the router to the receiver?
Explain.
P40. Consider Figure 3.58. Assuming TCP Reno is the protocol experiencing the
behavior shown above, answer the following questions. In all cases, you
should provide a short discussion justifying your answer.
a. Identify the intervals of time when TCP slow start is operating.
b. Identify the intervals of time when TCP congestion avoidance is
operating.
c. After the 16th transmission round, is segment loss detected by a triple
duplicate ACK or by a timeout?
d. After the 22nd transmission round, is segment loss detected by a triple
duplicate ACK or by a timeout?
e. What is the initial value of ssthresh at the first transmission round?
f. What is the value of ssthresh at the 18th transmission round?
g. What is the value of ssthresh at the 24th transmission round?
h. During what transmission round is the 70th segment sent?
i. Assuming a packet loss is detected after the 26th round by the receipt of a
triple duplicate ACK, what will be the values of the congestion window
size and of ssthresh?
45
Congestion window size (segments)
296
40
35
30
25
20
15
10
5
0
0
2
4
6
8
10 12 14 16 18 20 22 24 26
Transmission round
Figure 3.58
TCP window size as a function of time
PROBLEMS
P41.
P42.
P43.
P44.
j. Suppose TCP Tahoe is used (instead of TCP Reno), and assume that triple
duplicate ACKs are received at the 16th round. What are the ssthresh
and the congestion window size at the 19th round?
k. Again suppose TCP Tahoe is used, and there is a timeout event at 22nd
round. How many packets have been sent out from 17th round till 22nd
round, inclusive?
Refer to Figure 3.56, which illustrates the convergence of TCP’s AIMD
algorithm. Suppose that instead of a multiplicative decrease, TCP decreased
the window size by a constant amount. Would the resulting AIAD algorithm
converge to an equal share algorithm? Justify your answer using a diagram
similar to Figure 3.56.
In Section 3.5.4, we discussed the doubling of the timeout interval after a
timeout event. This mechanism is a form of congestion control. Why does
TCP need a window-based congestion-control mechanism (as studied in
Section 3.7) in addition to this doubling-timeout-interval mechanism?
Host A is sending an enormous file to Host B over a TCP connection.
Over this connection there is never any packet loss and the timers never
expire. Denote the transmission rate of the link connecting Host A to the
Internet by R bps. Suppose that the process in Host A is capable of sending
data into its TCP socket at a rate S bps, where S = 10 · R. Further suppose
that the TCP receive buffer is large enough to hold the entire file, and the
send buffer can hold only one percent of the file. What would prevent the
process in Host A from continuously passing data to its TCP socket at rate S
bps? TCP flow control? TCP congestion control? Or something else?
Elaborate.
Consider sending a large file from a host to another over a TCP connection
that has no loss.
a. Suppose TCP uses AIMD for its congestion control without slow start.
Assuming cwnd increases by 1 MSS every time a batch of ACKs is received
and assuming approximately constant round-trip times, how long does it take
for cwnd increase from 6 MSS to 12 MSS (assuming no loss events)?
b. What is the average throughout (in terms of MSS and RTT) for this connection up through time = 6 RTT?
P45. Recall the macroscopic description of TCP throughput. In the period of time
from when the connection’s rate varies from W/(2 · RTT) to W/RTT, only one
packet is lost (at the very end of the period).
a. Show that the loss rate (fraction of packets lost) is equal to
L = loss rate =
1
3 2
3
W + W
8
4
297
298
CHAPTER 3
•
TRANSPORT LAYER
b. Use the result above to show that if a connection has loss rate L, then its
average rate is approximately given by
⬇
1.22 ⭈ MSS
RTT 2L
P46. Consider that only a single TCP (Reno) connection uses one 10Mbps link
which does not buffer any data. Suppose that this link is the only congested
link between the sending and receiving hosts. Assume that the TCP sender
has a huge file to send to the receiver, and the receiver’s receive buffer is
much larger than the congestion window. We also make the following
assumptions: each TCP segment size is 1,500 bytes; the two-way propagation
delay of this connection is 150 msec; and this TCP connection is always in
congestion avoidance phase, that is, ignore slow start.
a. What is the maximum window size (in segments) that this TCP connection
can achieve?
b. What is the average window size (in segments) and average throughput (in
bps) of this TCP connection?
c. How long would it take for this TCP connection to reach its maximum
window again after recovering from a packet loss?
P47. Consider the scenario described in the previous problem. Suppose that the
10Mbps link can buffer a finite number of segments. Argue that in order for
the link to always be busy sending data, we would like to choose a buffer size
that is at least the product of the link speed C and the two-way propagation
delay between the sender and the receiver.
P48. Repeat Problem 43, but replacing the 10 Mbps link with a 10 Gbps link. Note
that in your answer to part c, you will realize that it takes a very long time for
the congestion window size to reach its maximum window size after recovering from a packet loss. Sketch a solution to solve this problem.
P49. Let T (measured by RTT) denote the time interval that a TCP connection
takes to increase its congestion window size from W/2 to W, where W is the
maximum congestion window size. Argue that T is a function of TCP’s average throughput.
P50. Consider a simplified TCP’s AIMD algorithm where the congestion window
size is measured in number of segments, not in bytes. In additive increase, the
congestion window size increases by one segment in each RTT. In multiplicative decrease, the congestion window size decreases by half (if the result is
not an integer, round down to the nearest integer). Suppose that two TCP
connections, C1 and C2, share a single congested link of speed 30 segments
per second. Assume that both C1 and C2 are in the congestion avoidance
PROBLEMS
phase. Connection C1’s RTT is 50 msec and connection C2’s RTT is
100 msec. Assume that when the data rate in the link exceeds the link’s
speed, all TCP connections experience data segment loss.
a. If both C1 and C2 at time t0 have a congestion window of 10 segments,
what are their congestion window sizes after 1000 msec?
b. In the long run, will these two connections get the same share of the bandwidth of the congested link? Explain.
P51. Consider the network described in the previous problem. Now suppose that
the two TCP connections, C1 and C2, have the same RTT of 100 msec. Suppose that at time t0, C1’s congestion window size is 15 segments but C2’s
congestion window size is 10 segments.
a. What are their congestion window sizes after 2200msec?
b. In the long run, will these two connections get about the same share of the
bandwidth of the congested link?
c. We say that two connections are synchronized, if both connections reach
their maximum window sizes at the same time and reach their minimum
window sizes at the same time. In the long run, will these two connections
get synchronized eventually? If so, what are their maximum window sizes?
d. Will this synchronization help to improve the utilization of the shared
link? Why? Sketch some idea to break this synchronization.
P52. Consider a modification to TCP’s congestion control algorithm. Instead of
additive increase, we can use multiplicative increase. A TCP sender increases
its window size by a small positive constant a (0 < a < 1) whenever it
receives a valid ACK. Find the functional relationship between loss rate L
and maximum congestion window W. Argue that for this modified TCP,
regardless of TCP’s average throughput, a TCP connection always spends the
same amount of time to increase its congestion window size from W/2 to W.
P53. In our discussion of TCP futures in Section 3.7, we noted that to achieve a
throughput of 10 Gbps, TCP could only tolerate a segment loss probability of
2 · 10-10 (or equivalently, one loss event for every 5,000,000,000 segments).
Show the derivation for the values of 2 · 10-10 (1 out of 5,000,000) for the
RTT and MSS values given in Section 3.7. If TCP needed to support a 100
Gbps connection, what would the tolerable loss be?
P54. In our discussion of TCP congestion control in Section 3.7, we implicitly
assumed that the TCP sender always had data to send. Consider now the case
that the TCP sender sends a large amount of data and then goes idle (since it
has no more data to send) at t1. TCP remains idle for a relatively long period of
time and then wants to send more data at t2. What are the advantages and disadvantages of having TCP use the cwnd and ssthresh values from t1 when
starting to send data at t2? What alternative would you recommend? Why?
299
300
CHAPTER 3
•
TRANSPORT LAYER
P55. In this problem we investigate whether either UDP or TCP provides a degree
of end-point authentication.
a. Consider a server that receives a request within a UDP packet and
responds to that request within a UDP packet (for example, as done by a
DNS server). If a client with IP address X spoofs its address with address
Y, where will the server send its response?
b. Suppose a server receives a SYN with IP source address Y, and after
responding with a SYNACK, receives an ACK with IP source address Y
with the correct acknowledgment number. Assuming the server chooses a
random initial sequence number and there is no “man-in-the-middle,” can
the server be certain that the client is indeed at Y (and not at some other
address X that is spoofing Y)?
P56. In this problem, we consider the delay introduced by the TCP slow-start
phase. Consider a client and a Web server directly connected by one link of
rate R. Suppose the client wants to retrieve an object whose size is exactly
equal to 15 S, where S is the maximum segment size (MSS). Denote the
round-trip time between client and server as RTT (assumed to be constant).
Ignoring protocol headers, determine the time to retrieve the object (including
TCP connection establishment) when
a. 4 S/R > S/R + RTT > 2S/R
b. S/R + RTT > 4 S/R
c. S/R > RTT.
Programming Assignments
Implementing a Reliable Transport Protocol
In this laboratory programming assignment, you will be writing the sending and
receiving transport-level code for implementing a simple reliable data transfer protocol. There are two versions of this lab, the alternating-bit-protocol version and the
GBN version. This lab should be fun—your implementation will differ very little
from what would be required in a real-world situation.
Since you probably don’t have standalone machines (with an OS that you can
modify), your code will have to execute in a simulated hardware/software environment. However, the programming interface provided to your routines—the
code that would call your entities from above and from below—is very close to
what is done in an actual UNIX environment. (Indeed, the software interfaces
described in this programming assignment are much more realistic than the infinite loop senders and receivers that many texts describe.) Stopping and starting
WIRESHARK LAB: EXPLORING UDP
timers are also simulated, and timer interrupts will cause your timer handling routine to be activated.
The full lab assignment, as well as code you will need to compile with your
own code, are available at this book’s Web site: http://www.awl.com/kurose-ross.
Wireshark Lab: Exploring TCP
In this lab, you’ll use your Web browser to access a file from a Web server. As in
earlier Wireshark labs, you’ll use Wireshark to capture the packets arriving at your
computer. Unlike earlier labs, you’ll also be able to download a Wireshark-readable
packet trace from the Web server from which you downloaded the file. In this server
trace, you’ll find the packets that were generated by your own access of the Web
server. You’ll analyze the client- and server-side traces to explore aspects of TCP. In
particular, you’ll evaluate the performance of the TCP connection between your
computer and the Web server. You’ll trace TCP’s window behavior, and infer packet
loss, retransmission, flow control and congestion control behavior, and estimated
roundtrip time.
As is the case with all Wireshark labs, the full description of this lab is available at this book’s Web site, http://www.awl.com/kurose-ross.
Wireshark Lab: Exploring UDP
In this short lab, you’ll do a packet capture and analysis of your favorite application
that uses UDP (for example, DNS or a multimedia application such as Skype). As we
learned in Section 3.3, UDP is a simple, no-frills transport protocol. In this lab, you’ll
investigate the header fields in the UDP segment as well as the checksum calculation.
As is the case with all Wireshark labs, the full description of this lab is available
at this book’s Web site, http://www.awl.com/kurose-ross.
301
AN INTERVIEW WITH...
Van Jacobson
Van Jacobson is a Research Fellow at PARC. Prior to that, he was
co-founder and Chief Scientist of Packet Design. Before that, he was
Chief Scientist at Cisco. Before joining Cisco, he was head of the
Network Research Group at Lawrence Berkeley National Laboratory
and taught at UC Berkeley and Stanford. Van received the ACM
SIGCOMM Award in 2001 for outstanding lifetime contribution to
the field of communication networks and the IEEE Kobayashi Award
in 2002 for “contributing to the understanding of network congestion
and developing congestion control mechanisms that enabled the successful scaling of the Internet”. He was elected to the U.S. National
Academy of Engineering in 2004.
Please describe one or two of the most exciting projects you have worked on during your
career. What were the biggest challenges?
School teaches us lots of ways to find answers. In every interesting problem I’ve worked
on, the challenge has been finding the right question. When Mike Karels and I started looking at TCP congestion, we spent months staring at protocol and packet traces asking “Why
is it failing?”. One day in Mike’s office, one of us said “The reason I can’t figure out why it
fails is because I don’t understand how it ever worked to begin with.” That turned out to be
the right question and it forced us to figure out the “ack clocking” that makes TCP work.
After that, the rest was easy.
More generally, where do you see the future of networking and the Internet?
For most people, the Web is the Internet. Networking geeks smile politely since we know
the Web is an application running over the Internet but what if they’re right? The Internet is
about enabling conversations between pairs of hosts. The Web is about distributed information production and consumption. “Information propagation” is a very general view of communication of which “pairwise conversation” is a tiny subset. We need to move into the
larger tent. Networking today deals with broadcast media (radios, PONs, etc.) by pretending
it’s a point-to-point wire. That’s massively inefficient. Terabits-per-second of data are being
exchanged all over the World via thumb drives or smart phones but we don’t know how to
treat that as “networking”. ISPs are busily setting up caches and CDNs to scalably distribute
video and audio. Caching is a necessary part of the solution but there's no part of today's
networking—from Information, Queuing or Traffic Theory down to the Internet protocol
specs—that tells us how to engineer and deploy it. I think and hope that over the next few
years, networking will evolve to embrace the much larger vision of communication that
underlies the Web.
302
What people inspired you professionally?
When I was in grad school, Richard Feynman visited and gave a colloquium. He talked
about a piece of Quantum theory that I’d been struggling with all semester and his explanation was so simple and lucid that what had been incomprehensible gibberish to me became
obvious and inevitable. That ability to see and convey the simplicity that underlies our complex world seems to me a rare and wonderful gift.
What are your recommendations for students who want careers in computer science and
networking?
It’s a wonderful field—computers and networking have probably had more impact on society than any invention since the book. Networking is fundamentally about connecting stuff,
and studying it helps you make intellectual connections: Ant foraging & Bee dances demonstrate protocol design better than RFCs, traffic jams or people leaving a packed stadium are
the essence of congestion, and students finding flights back to school in a post-Thanksgiving
blizzard are the core of dynamic routing. If you’re interested in lots of stuff and want to
have an impact, it’s hard to imagine a better field.
303
This page intentionally left blank
CHAPTER
4
The Network
Layer
We learned in the previous chapter that the transport layer provides various forms of
process-to-process communication by relying on the network layer’s host-to-host
communication service. We also learned that the transport layer does so without any
knowledge about how the network layer actually implements this service. So perhaps you’re now wondering, what’s under the hood of the host-to-host communication service, what makes it tick?
In this chapter, we’ll learn exactly how the network layer implements the hostto-host communication service. We’ll see that unlike the transport and application
layers, there is a piece of the network layer in each and every host and router in the
network. Because of this, network-layer protocols are among the most challenging
(and therefore among the most interesting!) in the protocol stack.
The network layer is also one of the most complex layers in the protocol stack,
and so we’ll have a lot of ground to cover here. We’ll begin our study with an
overview of the network layer and the services it can provide. We’ll then examine
two broad approaches towards structuring network-layer packet delivery—the datagram and the virtual-circuit model—and see the fundamental role that addressing
plays in delivering a packet to its destination host.
In this chapter, we’ll make an important distinction between the forwarding
and routing functions of the network layer. Forwarding involves the transfer of a
packet from an incoming link to an outgoing link within a single router. Routing
305
306
CHAPTER 4
•
THE NETWORK LAYER
involves all of a network’s routers, whose collective interactions via routing protocols determine the paths that packets take on their trips from source to destination
node. This will be an important distinction to keep in mind as you progress through
this chapter.
In order to deepen our understanding of packet forwarding, we’ll look “inside”
a router—at its hardware architecture and organization. We’ll then look at packet
forwarding in the Internet, along with the celebrated Internet Protocol (IP). We’ll
investigate network-layer addressing and the IPv4 datagram format. We’ll then
explore network address translation (NAT), datagram fragmentation, the Internet
Control Message Protocol (ICMP), and IPv6.
We’ll then turn our attention to the network layer’s routing function. We’ll see
that the job of a routing algorithm is to determine good paths (equivalently, routes)
from senders to receivers. We’ll first study the theory of routing algorithms, concentrating on the two most prevalent classes of algorithms: link-state and distancevector algorithms. Since the complexity of routing algorithms grows considerably
as the number of network routers increases, hierarchical routing approaches will
also be of interest. We’ll then see how theory is put into practice when we cover the
Internet’s intra-autonomous system routing protocols (RIP, OSPF, and IS-IS) and its
inter-autonomous system routing protocol, BGP. We’ll close this chapter with a discussion of broadcast and multicast routing.
In summary, this chapter has three major parts. The first part, Sections 4.1 and
4.2, covers network-layer functions and services. The second part, Sections 4.3 and
4.4, covers forwarding. Finally, the third part, Sections 4.5 through 4.7, covers
routing.
4.1 Introduction
Figure 4.1 shows a simple network with two hosts, H1 and H2, and several routers
on the path between H1 and H2. Suppose that H1 is sending information to H2, and
consider the role of the network layer in these hosts and in the intervening routers.
The network layer in H1 takes segments from the transport layer in H1, encapsulates each segment into a datagram (that is, a network-layer packet), and then sends
the datagrams to its nearby router, R1. At the receiving host, H2, the network layer
receives the datagrams from its nearby router R2, extracts the transport-layer segments, and delivers the segments up to the transport layer at H2. The primary role of
the routers is to forward datagrams from input links to output links. Note that the
routers in Figure 4.1 are shown with a truncated protocol stack, that is, with no
upper layers above the network layer, because (except for control purposes) routers
do not run application- and transport-layer protocols such as those we examined in
Chapters 2 and 3.
4.1
•
INTRODUCTION
307
National or
Global ISP
Mobile Network
Network
Data link
End system H1
Router R1
Application
Network
Transport
Data link
Network
Physical
Physical
Local or
Regional ISP
Data link
Physical
Network
Data link
Physical
Home Network
Network
Data link
Router R2
Physical
Network
Data link
Physical
End system H2
Application
Transport
Network
Data link
Enterprise Network
Figure 4.1
The network layer
Physical
308
CHAPTER 4
•
THE NETWORK LAYER
4.1.1 Forwarding and Routing
The role of the network layer is thus deceptively simple—to move packets from a
sending host to a receiving host. To do so, two important network-layer functions
can be identified:
• Forwarding. When a packet arrives at a router’s input link, the router must move
the packet to the appropriate output link. For example, a packet arriving from
Host H1 to Router R1 must be forwarded to the next router on a path to H2. In
Section 4.3, we’ll look inside a router and examine how a packet is actually forwarded from an input link to an output link within a router.
• Routing. The network layer must determine the route or path taken by packets as
they flow from a sender to a receiver. The algorithms that calculate these paths
are referred to as routing algorithms. A routing algorithm would determine, for
example, the path along which packets flow from H1 to H2.
The terms forwarding and routing are often used interchangeably by authors discussing the network layer. We’ll use these terms much more precisely in this book.
Forwarding refers to the router-local action of transferring a packet from an input link
interface to the appropriate output link interface. Routing refers to the network-wide
process that determines the end-to-end paths that packets take from source to destination. Using a driving analogy, consider the trip from Pennsylvania to Florida undertaken by our traveler back in Section 1.3.1. During this trip, our driver passes through
many interchanges en route to Florida. We can think of forwarding as the process of
getting through a single interchange: A car enters the interchange from one road and
determines which road it should take to leave the interchange. We can think of routing
as the process of planning the trip from Pennsylvania to Florida: Before embarking on
the trip, the driver has consulted a map and chosen one of many paths possible, with
each path consisting of a series of road segments connected at interchanges.
Every router has a forwarding table. A router forwards a packet by examining the value of a field in the arriving packet’s header, and then using this header
value to index into the router’s forwarding table. The value stored in the forwarding table entry for that header indicates the router’s outgoing link interface to
which that packet is to be forwarded. Depending on the network-layer protocol,
the header value could be the destination address of the packet or an indication of
the connection to which the packet belongs. Figure 4.2 provides an example. In
Figure 4.2, a packet with a header field value of 0111 arrives to a router. The
router indexes into its forwarding table and determines that the output link
interface for this packet is interface 2. The router then internally forwards the
packet to interface 2. In Section 4.3, we’ll look inside a router and examine the
forwarding function in much greater detail.
You might now be wondering how the forwarding tables in the routers are configured. This is a crucial issue, one that exposes the important interplay between
4.1
•
INTRODUCTION
Routing algorithm
Local forwarding table
header value output link
0100
0101
0111
1001
3
2
2
1
Value in arriving
packet’s header
1
0111
3
Figure 4.2
2
Routing algorithms determine values in forwarding tables
routing and forwarding. As shown in Figure 4.2, the routing algorithm determines
the values that are inserted into the routers’ forwarding tables. The routing algorithm
may be centralized (e.g., with an algorithm executing on a central site and downloading routing information to each of the routers) or decentralized (i.e., with a
piece of the distributed routing algorithm running in each router). In either case, a
router receives routing protocol messages, which are used to configure its forwarding table. The distinct and different purposes of the forwarding and routing functions can be further illustrated by considering the hypothetical (and unrealistic, but
technically feasible) case of a network in which all forwarding tables are configured
directly by human network operators physically present at the routers. In this case,
no routing protocols would be required! Of course, the human operators would need
to interact with each other to ensure that the forwarding tables were configured in
such a way that packets reached their intended destinations. It’s also likely that
human configuration would be more error-prone and much slower to respond to
changes in the network topology than a routing protocol. We’re thus fortunate that
all networks have both a forwarding and a routing function!
309
310
CHAPTER 4
•
THE NETWORK LAYER
While we’re on the topic of terminology, it’s worth mentioning two other terms
that are often used interchangeably, but that we will use more carefully. We’ll reserve
the term packet switch to mean a general packet-switching device that transfers a
packet from input link interface to output link interface, according to the value in a field
in the header of the packet. Some packet switches, called link-layer switches (examined in Chapter 5), base their forwarding decision on values in the fields of the linklayer frame; switches are thus referred to as link-layer (layer 2) devices. Other packet
switches, called routers, base their forwarding decision on the value in the networklayer field. Routers are thus network-layer (layer 3) devices, but must also implement
layer 2 protocols as well, since layer 3 devices require the services of layer 2 to implement their (layer 3) functionality. (To fully appreciate this important distinction, you
might want to review Section 1.5.2, where we discuss network-layer datagrams and
link-layer frames and their relationship.) To confuse matters, marketing literature often
refers to “layer 3 switches” for routers with Ethernet interfaces, but these are really
layer 3 devices. Since our focus in this chapter is on the network layer, we use the term
router in place of packet switch. We’ll even use the term router when talking about
packet switches in virtual-circuit networks (soon to be discussed).
Connection Setup
We just said that the network layer has two important functions, forwarding and routing. But we’ll soon see that in some computer networks there is actually a third important network-layer function, namely, connection setup. Recall from our study of TCP
that a three-way handshake is required before data can flow from sender to receiver.
This allows the sender and receiver to set up the needed state information (for example,
sequence number and initial flow-control window size). In an analogous manner, some
network-layer architectures—for example, ATM, frame relay, and MPLS (which we
will study in Section 5.8)––require the routers along the chosen path from source to
destination to handshake with each other in order to set up state before network-layer
data packets within a given source-to-destination connection can begin to flow. In the
network layer, this process is referred to as connection setup. We’ll examine connection setup in Section 4.2.
4.1.2 Network Service Models
Before delving into the network layer, let’s take the broader view and consider the different types of service that might be offered by the network layer. When the transport
layer at a sending host transmits a packet into the network (that is, passes it down to
the network layer at the sending host), can the transport layer rely on the network layer
to deliver the packet to the destination? When multiple packets are sent, will they be
delivered to the transport layer in the receiving host in the order in which they were
sent? Will the amount of time between the sending of two sequential packet transmissions be the same as the amount of time between their reception? Will the network
4.1
•
INTRODUCTION
provide any feedback about congestion in the network? What is the abstract view
(properties) of the channel connecting the transport layer in the sending and receiving
hosts? The answers to these questions and others are determined by the service model
provided by the network layer. The network service model defines the characteristics
of end-to-end transport of packets between sending and receiving end systems.
Let’s now consider some possible services that the network layer could provide.
In the sending host, when the transport layer passes a packet to the network layer,
specific services that could be provided by the network layer include:
• Guaranteed delivery. This service guarantees that the packet will eventually
arrive at its destination.
• Guaranteed delivery with bounded delay. This service not only guarantees delivery of the packet, but delivery within a specified host-to-host delay bound (for
example, within 100 msec).
Furthermore, the following services could be provided to a flow of packets between
a given source and destination:
• In-order packet delivery. This service guarantees that packets arrive at the destination in the order that they were sent.
• Guaranteed minimal bandwidth. This network-layer service emulates the behavior
of a transmission link of a specified bit rate (for example, 1 Mbps) between sending and receiving hosts. As long as the sending host transmits bits (as part of packets) at a rate below the specified bit rate, then no packet is lost and each packet
arrives within a prespecified host-to-host delay (for example, within 40 msec).
• Guaranteed maximum jitter. This service guarantees that the amount of time
between the transmission of two successive packets at the sender is equal to the
amount of time between their receipt at the destination (or that this spacing
changes by no more than some specified value).
• Security services. Using a secret session key known only by a source and destination host, the network layer in the source host could encrypt the payloads of
all datagrams being sent to the destination host. The network layer in the
destination host would then be responsible for decrypting the payloads. With
such a service, confidentiality would be provided to all transport-layer segments
(TCP and UDP) between the source and destination hosts. In addition to confidentiality, the network layer could provide data integrity and source authentication services.
This is only a partial list of services that a network layer could provide—there are
countless variations possible.
The Internet’s network layer provides a single service, known as best-effort
service. From Table 4.1, it might appear that best-effort service is a euphemism for
311
312
CHAPTER 4
•
THE NETWORK LAYER
Network
Architecture
Service
Model
Bandwidth
Guarantee
No-Loss
Guarantee
Internet
Best Effort
None
None
ATM
CBR
ATM
ABR
Guaranteed
constant rate
Guaranteed
minimum
Table 4.1
Ordering
Timing
Yes
Any order
possible
In order
Not
maintained
Maintained
None
In order
Not
maintained
Congestion
Indication
None
Congestion
will not occur
Congestion
indication
provided
Internet, ATM CBR, and ATM ABR service models
no service at all. With best-effort service, timing between packets is not guaranteed
to be preserved, packets are not guaranteed to be received in the order in which they
were sent, nor is the eventual delivery of transmitted packets guaranteed. Given this
definition, a network that delivered no packets to the destination would satisfy the
definition of best-effort delivery service. As we’ll discuss shortly, however, there
are sound reasons for such a minimalist network-layer service model.
Other network architectures have defined and implemented service models that
go beyond the Internet’s best-effort service. For example, the ATM network architecture [MFA Forum 2012, Black 1995] provides for multiple service models, meaning that different connections can be provided with different classes of service
within the same network. A discussion of how an ATM network provides such services is well beyond the scope of this book; our aim here is only to note that alternatives do exist to the Internet’s best-effort model. Two of the more important ATM
service models are constant bit rate and available bit rate service:
• Constant bit rate (CBR) ATM network service. This was the first ATM service
model to be standardized, reflecting early interest by the telephone companies in
ATM and the suitability of CBR service for carrying real-time, constant bit rate
audio and video traffic. The goal of CBR service is conceptually simple—to provide a flow of packets (known as cells in ATM terminology) with a virtual pipe
whose properties are the same as if a dedicated fixed-bandwidth transmission
link existed between sending and receiving hosts. With CBR service, a flow of
ATM cells is carried across the network in such a way that a cell’s end-to-end
delay, the variability in a cell’s end-to-end delay (that is, the jitter), and the fraction
of cells that are lost or delivered late are all guaranteed to be less than specified
values. These values are agreed upon by the sending host and the ATM network
when the CBR connection is first established.
4.2
•
VIRTUAL CIRCUIT AND DATAGRAM NETWORKS
• Available bit rate (ABR) ATM network service. With the Internet offering socalled best-effort service, ATM’s ABR might best be characterized as being a
slightly-better-than-best-effort service. As with the Internet service model,
cells may be lost under ABR service. Unlike in the Internet, however, cells
cannot be reordered (although they may be lost), and a minimum cell transmission rate (MCR) is guaranteed to a connection using ABR service. If the network has enough free resources at a given time, a sender may also be able to
send cells successfully at a higher rate than the MCR. Additionally, as we saw
in Section 3.6, ATM ABR service can provide feedback to the sender (in terms
of a congestion notification bit, or an explicit rate at which to send) that controls how the sender adjusts its rate between the MCR and an allowable peak
cell rate.
4.2 Virtual Circuit and Datagram Networks
Recall from Chapter 3 that a transport layer can offer applications connectionless
service or connection-oriented service between two processes. For example, the Internet’s transport layer provides each application a choice between two services: UDP, a
connectionless service; or TCP, a connection-oriented service. In a similar manner, a
network layer can provide connectionless service or connection service between two
hosts. Network-layer connection and connectionless services in many ways parallel
transport-layer connection-oriented and connectionless services. For example, a network-layer connection service begins with handshaking between the source and destination hosts; and a network-layer connectionless service does not have any
handshaking preliminaries.
Although the network-layer connection and connectionless services have some
parallels with transport-layer connection-oriented and connectionless services, there
are crucial differences:
• In the network layer, these services are host-to-host services provided by the network layer for the transport layer. In the transport layer these services are processto-process services provided by the transport layer for the application layer.
• In all major computer network architectures to date (Internet, ATM, frame relay,
and so on), the network layer provides either a host-to-host connectionless service or a host-to-host connection service, but not both. Computer networks that
provide only a connection service at the network layer are called virtual-circuit
(VC) networks; computer networks that provide only a connectionless service
at the network layer are called datagram networks.
• The implementations of connection-oriented service in the transport layer and
the connection service in the network layer are fundamentally different. We saw
in the previous chapter that the transport-layer connection-oriented service is
313
314
CHAPTER 4
•
THE NETWORK LAYER
implemented at the edge of the network in the end systems; we’ll see shortly that
the network-layer connection service is implemented in the routers in the network core as well as in the end systems.
Virtual-circuit and datagram networks are two fundamental classes of computer networks. They use very different information in making their forwarding decisions.
Let’s now take a closer look at their implementations.
4.2.1 Virtual-Circuit Networks
While the Internet is a datagram network, many alternative network architectures—
including those of ATM and frame relay—are virtual-circuit networks and, therefore, use connections at the network layer. These network-layer connections are
called virtual circuits (VCs). Let’s now consider how a VC service can be implemented in a computer network.
A VC consists of (1) a path (that is, a series of links and routers) between the
source and destination hosts, (2) VC numbers, one number for each link along the
path, and (3) entries in the forwarding table in each router along the path. A packet
belonging to a virtual circuit will carry a VC number in its header. Because a virtual
circuit may have a different VC number on each link, each intervening router must
replace the VC number of each traversing packet with a new VC number. The new
VC number is obtained from the forwarding table.
To illustrate the concept, consider the network shown in Figure 4.3. The numbers
next to the links of R1 in Figure 4.3 are the link interface numbers. Suppose now that
Host A requests that the network establish a VC between itself and Host B. Suppose
also that the network chooses the path A-R1-R2-B and assigns VC numbers 12, 22,
and 32 to the three links in this path for this virtual circuit. In this case, when a packet
in this VC leaves Host A, the value in the VC number field in the packet header is 12;
when it leaves R1, the value is 22; and when it leaves R2, the value is 32.
How does the router determine the replacement VC number for a packet traversing the router? For a VC network, each router’s forwarding table includes VC
A
R1
1
Figure 4.3
R2
2
1
B
2
3
3
R3
R4
A simple virtual circuit network
4.2
•
VIRTUAL CIRCUIT AND DATAGRAM NETWORKS
number translation; for example, the forwarding table in R1 might look something
like this:
Incoming Interface
Incoming VC #
Outgoing Interface
Outgoing VC #
1
12
2
22
2
63
1
18
3
7
2
17
1
97
3
87
...
...
...
...
Whenever a new VC is established across a router, an entry is added to the forwarding table. Similarly, whenever a VC terminates, the appropriate entries in each table
along its path are removed.
You might be wondering why a packet doesn’t just keep the same VC number
on each of the links along its route. The answer is twofold. First, replacing the number from link to link reduces the length of the VC field in the packet header. Second,
and more importantly, VC setup is considerably simplified by permitting a different
VC number at each link along the path of the VC. Specifically, with multiple VC
numbers, each link in the path can choose a VC number independently of the VC
numbers chosen at other links along the path. If a common VC number were required
for all links along the path, the routers would have to exchange and process a substantial number of messages to agree on a common VC number (e.g., one that is not
being used by any other existing VC at these routers) to be used for a connection.
In a VC network, the network’s routers must maintain connection state information for the ongoing connections. Specifically, each time a new connection is
established across a router, a new connection entry must be added to the router’s forwarding table; and each time a connection is released, an entry must be removed
from the table. Note that even if there is no VC-number translation, it is still necessary to maintain connection state information that associates VC numbers with output interface numbers. The issue of whether or not a router maintains connection
state information for each ongoing connection is a crucial one—one that we’ll return
to repeatedly in this book.
There are three identifiable phases in a virtual circuit:
• VC setup. During the setup phase, the sending transport layer contacts the network layer, specifies the receiver’s address, and waits for the network to set up
the VC. The network layer determines the path between sender and receiver, that
is, the series of links and routers through which all packets of the VC will travel.
The network layer also determines the VC number for each link along the path.
Finally, the network layer adds an entry in the forwarding table in each router
315
316
CHAPTER 4
•
THE NETWORK LAYER
along the path. During VC setup, the network layer may also reserve resources
(for example, bandwidth) along the path of the VC.
• Data transfer. As shown in Figure 4.4, once the VC has been established, packets can begin to flow along the VC.
• VC teardown. This is initiated when the sender (or receiver) informs the network
layer of its desire to terminate the VC. The network layer will then typically
inform the end system on the other side of the network of the call termination
and update the forwarding tables in each of the packet routers on the path to indicate that the VC no longer exists.
There is a subtle but important distinction between VC setup at the network
layer and connection setup at the transport layer (for example, the TCP three-way
handshake we studied in Chapter 3). Connection setup at the transport layer
involves only the two end systems. During transport-layer connection setup, the
two end systems alone determine the parameters (for example, initial sequence
number and flow-control window size) of their transport-layer connection.
Although the two end systems are aware of the transport-layer connection, the
routers within the network are completely oblivious to it. On the other hand, with
a VC network layer, routers along the path between the two end systems are
involved in VC setup, and each router is fully aware of all the VCs passing
through it.
The messages that the end systems send into the network to initiate or terminate a
VC, and the messages passed between the routers to set up the VC (that is, to modify
connection state in router tables) are known as signaling messages, and the protocols
Application
Transport
Network
Application
4. Call connected
3. Accept call
5. Data flow
begins
6. Receive
data
Transport
Network
Data link
Data link
Physical
Physical
1. Initiate call
2. Incoming call
Figure 4.4
Virtual-circuit setup
4.2
•
VIRTUAL CIRCUIT AND DATAGRAM NETWORKS
used to exchange these messages are often referred to as signaling protocols. VC setup
is shown pictorially in Figure 4.4. We’ll not cover VC signaling protocols in this book;
see [Black 1997] for a general discussion of signaling in connection-oriented networks
and [ITU-T Q.2931 1995] for the specification of ATM’s Q.2931 signaling protocol.
4.2.2 Datagram Networks
In a datagram network, each time an end system wants to send a packet, it stamps
the packet with the address of the destination end system and then pops the packet
into the network. As shown in Figure 4.5, there is no VC setup and routers do not
maintain any VC state information (because there are no VCs!).
As a packet is transmitted from source to destination, it passes through a series
of routers. Each of these routers uses the packet’s destination address to forward the
packet. Specifically, each router has a forwarding table that maps destination
addresses to link interfaces; when a packet arrives at the router, the router uses the
packet’s destination address to look up the appropriate output link interface in the
forwarding table. The router then intentionally forwards the packet to that output
link interface.
To get some further insight into the lookup operation, let’s look at a specific
example. Suppose that all destination addresses are 32 bits (which just happens to
be the length of the destination address in an IP datagram). A brute-force implementation of the forwarding table would have one entry for every possible destination
address. Since there are more than 4 billion possible addresses, this option is totally
out of the question.
Application
Application
Transport
Transport
Network
Data link
1. Send
data
Physical
Figure 4.5
Datagram network
2. Receive
data
Network
Data link
Physical
317
318
CHAPTER 4
•
THE NETWORK LAYER
Now let’s further suppose that our router has four links, numbered 0 through 3,
and that packets are to be forwarded to the link interfaces as follows:
Destination Address Range
Link Interface
11001000 00010111 00010000 00000000
through
11001000 00010111 00010111 11111111
0
11001000 00010111 00011000 00000000
through
11001000 00010111 00011000 11111111
1
11001000 00010111 00011001 00000000
through
11001000 00010111 00011111 11111111
2
otherwise
3
Clearly, for this example, it is not necessary to have 4 billion entries in the router’s
forwarding table. We could, for example, have the following forwarding table with
just four entries:
Prefix Match
11001000 00010111 00010
11001000 00010111 00011000
11001000 00010111 00011
otherwise
Link Interface
0
1
2
3
With this style of forwarding table, the router matches a prefix of the packet’s destination address with the entries in the table; if there’s a match, the router forwards
the packet to a link associated with the match. For example, suppose the packet’s
destination address is 11001000 00010111 00010110 10100001; because the 21-bit
prefix of this address matches the first entry in the table, the router forwards the
packet to link interface 0. If a prefix doesn’t match any of the first three entries, then
the router forwards the packet to interface 3. Although this sounds simple enough,
there’s an important subtlety here. You may have noticed that it is possible for a destination address to match more than one entry. For example, the first 24 bits of the
address 11001000 00010111 00011000 10101010 match the second entry in the
table, and the first 21 bits of the address match the third entry in the table. When
there are multiple matches, the router uses the longest prefix matching rule; that
is, it finds the longest matching entry in the table and forwards the packet to the link
4.2
•
VIRTUAL CIRCUIT AND DATAGRAM NETWORKS
interface associated with the longest prefix match. We’ll see exactly why this
longest prefix-matching rule is used when we study Internet addressing in more
detail in Section 4.4.
Although routers in datagram networks maintain no connection state information, they nevertheless maintain forwarding state information in their forwarding
tables. However, the time scale at which this forwarding state information changes
is relatively slow. Indeed, in a datagram network the forwarding tables are modified
by the routing algorithms, which typically update a forwarding table every one-tofive minutes or so. In a VC network, a forwarding table in a router is modified
whenever a new connection is set up through the router or whenever an existing
connection through the router is torn down. This could easily happen at a microsecond timescale in a backbone, tier-1 router.
Because forwarding tables in datagram networks can be modified at any time, a
series of packets sent from one end system to another may follow different paths
through the network and may arrive out of order. [Paxson 1997] and [Jaiswal 2003]
present interesting measurement studies of packet reordering and other phenomena
in the public Internet.
4.2.3 Origins of VC and Datagram Networks
The evolution of datagram and VC networks reflects their origins. The notion of a
virtual circuit as a central organizing principle has its roots in the telephony world,
which uses real circuits. With call setup and per-call state being maintained at the
routers within the network, a VC network is arguably more complex than a datagram network (although see [Molinero-Fernandez 2002] for an interesting comparison of the complexity of circuit- versus packet-switched networks). This, too, is in
keeping with its telephony heritage. Telephone networks, by necessity, had their
complexity within the network, since they were connecting dumb end-system
devices such as rotary telephones. (For those too young to know, a rotary phone is
an analog telephone with no buttons—only a dial.)
The Internet as a datagram network, on the other hand, grew out of the need to
connect computers together. Given more sophisticated end-system devices, the
Internet architects chose to make the network-layer service model as simple as possible. As we have already seen in Chapters 2 and 3, additional functionality (for
example, in-order delivery, reliable data transfer, congestion control, and DNS name
resolution) is then implemented at a higher layer, in the end systems. This inverts
the model of the telephone network, with some interesting consequences:
• Since the resulting Internet network-layer service model makes minimal (no!)
service guarantees, it imposes minimal requirements on the network layer. This
makes it easier to interconnect networks that use very different link-layer technologies (for example, satellite, Ethernet, fiber, or radio) that have very different
transmission rates and loss characteristics. We will address the interconnection
of IP networks in detail in Section 4.4.
319
320
CHAPTER 4
•
THE NETWORK LAYER
• As we saw in Chapter 2, applications such as e-mail, the Web, and even some
network infrastructure services such as the DNS are implemented in hosts
(servers) at the network edge. The ability to add a new service simply by attaching a host to the network and defining a new application-layer protocol (such as
HTTP) has allowed new Internet applications such as the Web to be deployed in
a remarkably short period of time.
4.3
What’s Inside a Router?
Now that we’ve overviewed the network layer’s services and functions, let’s turn
our attention to its forwarding function—the actual transfer of packets from a
router’s incoming links to the appropriate outgoing links at that router. We
already took a brief look at a few aspects of forwarding in Section 4.2, namely,
addressing and longest prefix matching. We mention here in passing that the terms
forwarding and switching are often used interchangeably by computer-networking
researchers and practitioners; we’ll use both terms interchangeably in this
textbook as well.
A high-level view of a generic router architecture is shown in Figure 4.6. Four
router components can be identified:
• Input ports. An input port performs several key functions. It performs the
physical layer function of terminating an incoming physical link at a router;
this is shown in the leftmost box of the input port and the rightmost box of the
output port in Figure 4.6. An input port also performs link-layer functions
needed to interoperate with the link layer at the other side of the incoming
link; this is represented by the middle boxes in the input and output ports. Perhaps most crucially, the lookup function is also performed at the input port;
this will occur in the rightmost box of the input port. It is here that the forwarding table is consulted to determine the router output port to which an
arriving packet will be forwarded via the switching fabric. Control packets
(for example, packets carrying routing protocol information) are forwarded
from an input port to the routing processor. Note that the term port here—
referring to the physical input and output router interfaces—is distinctly
different from the software ports associated with network applications and
sockets discussed in Chapters 2 and 3.
• Switching fabric. The switching fabric connects the router’s input ports to its
output ports. This switching fabric is completely contained within the router—
a network inside of a network router!
• Output ports. An output port stores packets received from the switching fabric
and transmits these packets on the outgoing link by performing the necessary
link-layer and physical-layer functions. When a link is bidirectional (that is,
4.3
Routing, management
control plane (software)
•
WHAT’S INSIDE A ROUTER?
Routing
processor
Forwarding
data plane (hardware)
Input port
Input port
Figure 4.6
Output port
Switch
fabric
Output port
Router architecture
carries traffic in both directions), an output port will typically be paired with the
input port for that link on the same line card (a printed circuit board containing
one or more input ports, which is connected to the switching fabric).
• Routing processor. The routing processor executes the routing protocols (which
we’ll study in Section 4.6), maintains routing tables and attached link state information, and computes the forwarding table for the router. It also performs the
network management functions that we’ll study in Chapter 9.
Recall that in Section 4.1.1 we distinguished between a router’s forwarding and
routing functions. A router’s input ports, output ports, and switching fabric
together implement the forwarding function and are almost always implemented
in hardware, as shown in Figure 4.6. These forwarding functions are sometimes
collectively referred to as the router forwarding plane. To appreciate why a
hardware implementation is needed, consider that with a 10 Gbps input link and a
64-byte IP datagram, the input port has only 51.2 ns to process the datagram
before another datagram may arrive. If N ports are combined on a line card (as is
often done in practice), the datagram-processing pipeline must operate N times
faster—far too fast for software implementation. Forwarding plane hardware can
be implemented either using a router vendor’s own hardware designs, or constructed using purchased merchant-silicon chips (e.g., as sold by companies such
as Intel and Broadcom).
While the forwarding plane operates at the nanosecond time scale, a router’s
control functions—executing the routing protocols, responding to attached links that
321
322
CHAPTER 4
•
THE NETWORK LAYER
go up or down, and performing management functions such as those we’ll study in
Chapter 9—operate at the millisecond or second timescale. These router control
plane functions are usually implemented in software and execute on the routing
processor (typically a traditional CPU).
Before delving into the details of a router’s control and data plane, let’s return to
our analogy of Section 4.1.1, where packet forwarding was compared to cars entering
and leaving an interchange. Let’s suppose that the interchange is a roundabout, and that
before a car enters the roundabout, a bit of processing is required—the car stops at an
entry station and indicates its final destination (not at the local roundabout, but the ultimate destination of its journey). An attendant at the entry station looks up the final destination, determines the roundabout exit that leads to that final destination, and tells the
driver which roundabout exit to take. The car enters the roundabout (which may be
filled with other cars entering from other input roads and heading to other roundabout
exits) and eventually leaves at the prescribed roundabout exit ramp, where it may
encounter other cars leaving the roundabout at that exit.
We can recognize the principal router components in Figure 4.6 in this analogy—the entry road and entry station correspond to the input port (with a lookup
function to determine to local outgoing port); the roundabout corresponds to the
switch fabric; and the roundabout exit road corresponds to the output port. With
this analogy, it’s instructive to consider where bottlenecks might occur. What happens if cars arrive blazingly fast (for example, the roundabout is in Germany or
Italy!) but the station attendant is slow? How fast must the attendant work to ensure
there’s no backup on an entry road? Even with a blazingly fast attendant, what happens if cars traverse the roundabout slowly—can backups still occur? And what
happens if most of the entering cars all want to leave the roundabout at the same
exit ramp—can backups occur at the exit ramp or elsewhere? How should the
roundabout operate if we want to assign priorities to different cars, or block certain
cars from entering the roundabout in the first place? These are all analogous to critical questions faced by router and switch designers.
In the following subsections, we’ll look at router functions in more detail. [Iyer
2008, Chao 2001; Chuang 2005; Turner 1988; McKeown 1997a; Partridge 1998]
provide a discussion of specific router architectures. For concreteness, the ensuing
discussion assumes a datagram network in which forwarding decisions are based
on the packet’s destination address (rather than a VC number in a virtual-circuit
network). However, the concepts and techniques are quite similar for a virtualcircuit network.
4.3.1 Input Processing
A more detailed view of input processing is given in Figure 4.7. As discussed above,
the input port’s line termination function and link-layer processing implement the
physical and link layers for that individual input link. The lookup performed in the
input port is central to the router’s operation—it is here that the router uses the forwarding table to look up the output port to which an arriving packet will be
4.3
•
WHAT’S INSIDE A ROUTER?
CASE HISTORY
CISCO SYSTEMS: DOMINATING THE NETWORK CORE
As of this writing 2012, Cisco employs more than 65,000 people. How did this
gorilla of a networking company come to be? It all started in 1984 in the living room
of a Silicon Valley apartment.
Len Bosak and his wife Sandy Lerner were working at Stanford University when they
had the idea to build and sell Internet routers to research and academic institutions, the
primary adopters of the Internet at that time. Sandy Lerner came up with the name Cisco
(an abbreviation for San Francisco), and she also designed the company’s bridge logo.
Corporate headquarters was their living room, and they financed the project with credit
cards and moonlighting consulting jobs. At the end of 1986, Cisco’s revenues reached
$250,000 a month. At the end of 1987, Cisco succeeded in attracting venture capital—
$2 million from Sequoia Capital in exchange for one-third of the company. Over the next
few years, Cisco continued to grow and grab more and more market share. At the same
time, relations between Bosak/Lerner and Cisco management became strained. Cisco
went public in 1990; in the same year Lerner and Bosak left the company.
Over the years, Cisco has expanded well beyond the router market, selling security,
wireless caching, Ethernet switch, datacenter infrastructure, video conferencing, and
voice-over IP products and services. However, Cisco is facing increased international
competition, including from Huawei, a rapidly growing Chinese network-gear company. Other sources of competition for Cisco in the router and switched Ethernet space
include Alcatel-Lucent and Juniper.
forwarded via the switching fabric. The forwarding table is computed and updated
by the routing processor, with a shadow copy typically stored at each input port. The
forwarding table is copied from the routing processor to the line cards over a separate bus (e.g., a PCI bus) indicated by the dashed line from the routing processor to
the input line cards in Figure 4.6. With a shadow copy, forwarding decisions can be
made locally, at each input port, without invoking the centralized routing processor
on a per-packet basis and thus avoiding a centralized processing bottleneck.
Given the existence of a forwarding table, lookup is conceptually simple—we just
search through the forwarding table looking for the longest prefix match, as described
Line
termination
Figure 4.7
Data link
processing
(protocol,
decapsulation)
Input port processing
Lookup, fowarding,
queuing
Switch
fabric
323
324
CHAPTER 4
•
THE NETWORK LAYER
in Section 4.2.2. But at Gigabit transmission rates, this lookup must be performed in
nanoseconds (recall our earlier example of a 10 Gbps link and a 64-byte IP datagram).
Thus, not only must lookup be performed in hardware, but techniques beyond a simple
linear search through a large table are needed; surveys of fast lookup algorithms can be
found in [Gupta 2001, Ruiz-Sanchez 2001]. Special attention must also be paid to memory access times, resulting in designs with embedded on-chip DRAM and faster SRAM
(used as a DRAM cache) memories. Ternary Content Address Memories (TCAMs) are
also often used for lookup. With a TCAM, a 32-bit IP address is presented to the memory, which returns the content of the forwarding table entry for that address in essentially constant time. The Cisco 8500 has a 64K CAM for each input port.
Once a packet’s output port has been determined via the lookup, the packet can
be sent into the switching fabric. In some designs, a packet may be temporarily
blocked from entering the switching fabric if packets from other input ports are currently using the fabric. A blocked packet will be queued at the input port and then
scheduled to cross the fabric at a later point in time. We’ll take a closer look at the
blocking, queuing, and scheduling of packets (at both input ports and output ports)
in Section 4.3.4. Although “lookup” is arguably the most important action in input
port processing, many other actions must be taken: (1) physical- and link-layer processing must occur, as discussed above; (2) the packet’s version number, checksum
and time-to-live field—all of which we’ll study in Section 4.4.1—must be checked
and the latter two fields rewritten; and (3) counters used for network management
(such as the number of IP datagrams received) must be updated.
Let’s close our discussion of input port processing by noting that the input port
steps of looking up an IP address (“match”) then sending the packet into the switching
fabric (“action”) is a specific case of a more general “match plus action” abstraction
that is performed in many networked devices, not just routers. In link-layer switches
(covered in Chapter 5), link-layer destination addresses are looked up and several
actions may be taken in addition to sending the frame into the switching fabric towards
the output port. In firewalls (covered in Chapter 8)—devices that filter out selected
incoming packets—an incoming packet whose header matches a given criteria (e.g., a
combination of source/destination IP addresses and transport-layer port numbers) may
be prevented from being forwarded (action). In a network address translator (NAT, covered in Section 4.4), an incoming packet whose transport-layer port number matches a
given value will have its port number rewritten before forwarding (action). Thus, the
“match plus action” abstraction is both powerful and prevalent in network devices.
4.3.2 Switching
The switching fabric is at the very heart of a router, as it is through this fabric that
the packets are actually switched (that is, forwarded) from an input port to an output
port. Switching can be accomplished in a number of ways, as shown in Figure 4.8:
• Switching via memory. The simplest, earliest routers were traditional computers,
with switching between input and output ports being done under direct control of
4.3
Memory
•
WHAT’S INSIDE A ROUTER?
325
Crossbar
A
B
X
A
Y
B
Z
C
Memory
C
X
Bus
A
X
B
Y
C
Z
Key:
Input port
Figure 4.8
Output port
Three switching techniques
the CPU (routing processor). Input and output ports functioned as traditional I/O
devices in a traditional operating system. An input port with an arriving packet
first signaled the routing processor via an interrupt. The packet was then copied
from the input port into processor memory. The routing processor then extracted
the destination address from the header, looked up the appropriate output port in
the forwarding table, and copied the packet to the output port’s buffers. In this
scenario, if the memory bandwidth is such that B packets per second can be written into, or read from, memory, then the overall forwarding throughput (the total
rate at which packets are transferred from input ports to output ports) must be
less than B/2. Note also that two packets cannot be forwarded at the same time,
even if they have different destination ports, since only one memory read/write
over the shared system bus can be done at a time.
Many modern routers switch via memory. A major difference from early routers,
however, is that the lookup of the destination address and the storing of the packet
into the appropriate memory location are performed by processing on the input
line cards. In some ways, routers that switch via memory look very much like
shared-memory multiprocessors, with the processing on a line card switching
(writing) packets into the memory of the appropriate output port. Cisco’s Catalyst
8500 series switches [Cisco 8500 2012] forward packets via a shared memory.
Y
Z
326
CHAPTER 4
•
THE NETWORK LAYER
• Switching via a bus. In this approach, an input port transfers a packet directly to the
output port over a shared bus, without intervention by the routing processor. This is
typically done by having the input port pre-pend a switch-internal label (header) to
the packet indicating the local output port to which this packet is being transferred
and transmitting the packet onto the bus. The packet is received by all output ports,
but only the port that matches the label will keep the packet. The label is then
removed at the output port, as this label is only used within the switch to cross the
bus. If multiple packets arrive to the router at the same time, each at a different input
port, all but one must wait since only one packet can cross the bus at a time. Because
every packet must cross the single bus, the switching speed of the router is limited
to the bus speed; in our roundabout analogy, this is as if the roundabout could only
contain one car at a time. Nonetheless, switching via a bus is often sufficient for
routers that operate in small local area and enterprise networks. The Cisco 5600
[Cisco Switches 2012] switches packets over a 32 Gbps backplane bus.
• Switching via an interconnection network. One way to overcome the bandwidth
limitation of a single, shared bus is to use a more sophisticated interconnection network, such as those that have been used in the past to interconnect processors in a
multiprocessor computer architecture. A crossbar switch is an interconnection network consisting of 2N buses that connect N input ports to N output ports, as shown
in Figure 4.8. Each vertical bus intersects each horizontal bus at a crosspoint, which
can be opened or closed at any time by the switch fabric controller (whose logic is
part of the switching fabric itself). When a packet arrives from port A and needs to
be forwarded to port Y, the switch controller closes the crosspoint at the intersection
of busses A and Y, and port A then sends the packet onto its bus, which is picked up
(only) by bus Y. Note that a packet from port B can be forwarded to port X at the
same time, since the A-to-Y and B-to-X packets use different input and output
busses. Thus, unlike the previous two switching approaches, crossbar networks are
capable of forwarding multiple packets in parallel. However, if two packets from
two different input ports are destined to the same output port, then one will have to
wait at the input, since only one packet can be sent over any given bus at a time.
More sophisticated interconnection networks use multiple stages of switching
elements to allow packets from different input ports to proceed towards the same
output port at the same time through the switching fabric. See [Tobagi 1990] for
a survey of switch architectures. Cisco 12000 family switches [Cisco 12000
2012] use an interconnection network.
4.3.3 Output Processing
Output port processing, shown in Figure 4.9, takes packets that have been stored in
the output port’s memory and transmits them over the output link. This includes
selecting and de-queueing packets for transmission, and performing the needed linklayer and physical-layer transmission functions.
4.3
Queuing (buffer
management)
Switch
fabric
Figure 4.9
Data link
processing
(protocol,
encapsulation)
•
WHAT’S INSIDE A ROUTER?
Line
termination
Output port processing
4.3.4 Where Does Queueing Occur?
If we consider input and output port functionality and the configurations shown in
Figure 4.8, it’s clear that packet queues may form at both the input ports and the output ports, just as we identified cases where cars may wait at the inputs and outputs of
the traffic intersection in our roundabout analogy. The location and extent of queueing
(either at the input port queues or the output port queues) will depend on the traffic
load, the relative speed of the switching fabric, and the line speed. Let’s now consider
these queues in a bit more detail, since as these queues grow large, the router’s memory can eventually be exhausted and packet loss will occur when no memory is available to store arriving packets. Recall that in our earlier discussions, we said that
packets were “lost within the network” or “dropped at a router.” It is here, at these
queues within a router, where such packets are actually dropped and lost.
Suppose that the input and output line speeds (transmission rates) all have an
identical transmission rate of Rline packets per second, and that there are N input
ports and N output ports. To further simplify the discussion, let’s assume that all
packets have the same fixed length, and the packets arrive to input ports in a synchronous manner. That is, the time to send a packet on any link is equal to the time
to receive a packet on any link, and during such an interval of time, either zero or
one packet can arrive on an input link. Define the switching fabric transfer rate
Rswitch as the rate at which packets can be moved from input port to output port. If
Rswitch is N times faster than Rline, then only negligible queuing will occur at the
input ports. This is because even in the worst case, where all N input lines are
receiving packets, and all packets are to be forwarded to the same output port, each
batch of N packets (one packet per input port) can be cleared through the switch fabric before the next batch arrives.
But what can happen at the output ports? Let’s suppose that Rswitch is still N
times faster than Rline. Once again, packets arriving at each of the N input ports
are destined to the same output port. In this case, in the time it takes to send a single
packet onto the outgoing link, N new packets will arrive at this output port. Since
the output port can transmit only a single packet in a unit of time (the packet transmission time), the N arriving packets will have to queue (wait) for transmission over
the outgoing link. Then N more packets can possibly arrive in the time it takes to
327
328
CHAPTER 4
•
THE NETWORK LAYER
transmit just one of the N packets that had just previously been queued. And so on.
Eventually, the number of queued packets can grow large enough to exhaust available memory at the output port, in which case packets are dropped.
Output port queuing is illustrated in Figure 4.10. At time t, a packet has arrived at
each of the incoming input ports, each destined for the uppermost outgoing port.
Assuming identical line speeds and a switch operating at three times the line speed,
one time unit later (that is, in the time needed to receive or send a packet), all three
original packets have been transferred to the outgoing port and are queued awaiting
transmission. In the next time unit, one of these three packets will have been transmitted over the outgoing link. In our example, two new packets have arrived at the incoming side of the switch; one of these packets is destined for this uppermost output port.
Given that router buffers are needed to absorb the fluctuations in traffic load, the
natural question to ask is how much buffering is required. For many years, the rule of
thumb [RFC 3439] for buffer sizing was that the amount of buffering (B) should be
equal to an average round-trip time (RTT, say 250 msec) times the link capacity (C).
This result is based on an analysis of the queueing dynamics of a relatively small number of TCP flows [Villamizar 1994]. Thus, a 10 Gbps link with an RTT of 250 msec
would need an amount of buffering equal to B = RTT · C = 2.5 Gbits of buffers. Recent
Output port contention at time t
Switch
fabric
One packet time later
Switch
fabric
Figure 4.10
Output port queuing
4.3
•
WHAT’S INSIDE A ROUTER?
theoretical and experimental efforts [Appenzeller 2004], however, suggest that when
there are a large number of TCP flows (N) passing through a link, the amount of buffer—
ing needed is B = RTT ⭈ C/√N . With a large number of flows typically passing through
large backbone router links (see, e.g., [Fraleigh 2003]), the value of N can be large, with
the decrease in needed buffer size becoming quite significant. [Appenzellar 2004; Wischik 2005; Beheshti 2008] provide very readable discussions of the buffer sizing problem from a theoretical, implementation, and operational standpoint.
A consequence of output port queuing is that a packet scheduler at the output
port must choose one packet among those queued for transmission. This selection
might be done on a simple basis, such as first-come-first-served (FCFS) scheduling,
or a more sophisticated scheduling discipline such as weighted fair queuing (WFQ),
which shares the outgoing link fairly among the different end-to-end connections
that have packets queued for transmission. Packet scheduling plays a crucial role in
providing quality-of-service guarantees. We’ll thus cover packet scheduling extensively in Chapter 7. A discussion of output port packet scheduling disciplines is
[Cisco Queue 2012].
Similarly, if there is not enough memory to buffer an incoming packet, a decision
must be made to either drop the arriving packet (a policy known as drop-tail) or
remove one or more already-queued packets to make room for the newly arrived
packet. In some cases, it may be advantageous to drop (or mark the header of) a packet
before the buffer is full in order to provide a congestion signal to the sender. A number
of packet-dropping and -marking policies (which collectively have become known as
active queue management (AQM) algorithms) have been proposed and analyzed
[Labrador 1999, Hollot 2002]. One of the most widely studied and implemented AQM
algorithms is the Random Early Detection (RED) algorithm. Under RED, a
weighted average is maintained for the length of the output queue. If the average
queue length is less than a minimum threshold, minth, when a packet arrives, the
packet is admitted to the queue. Conversely, if the queue is full or the average queue
length is greater than a maximum threshold, maxth, when a packet arrives, the packet
is marked or dropped. Finally, if the packet arrives to find an average queue length in
the interval [minth, maxth], the packet is marked or dropped with a probability that is
typically some function of the average queue length, minth, and maxth. A number of
probabilistic marking/dropping functions have been proposed, and various versions of
RED have been analytically modeled, simulated, and/or implemented. [Christiansen
2001] and [Floyd 2012] provide overviews and pointers to additional reading.
If the switch fabric is not fast enough (relative to the input line speeds) to transfer
all arriving packets through the fabric without delay, then packet queuing can also
occur at the input ports, as packets must join input port queues to wait their turn to be
transferred through the switching fabric to the output port. To illustrate an important
consequence of this queuing, consider a crossbar switching fabric and suppose that
(1) all link speeds are identical, (2) that one packet can be transferred from any one
input port to a given output port in the same amount of time it takes for a packet to be
received on an input link, and (3) packets are moved from a given input queue to their
329
330
CHAPTER 4
•
THE NETWORK LAYER
desired output queue in an FCFS manner. Multiple packets can be transferred in parallel, as long as their output ports are different. However, if two packets at the front of
two input queues are destined for the same output queue, then one of the packets will
be blocked and must wait at the input queue—the switching fabric can transfer only
one packet to a given output port at a time.
Figure 4.11 shows an example in which two packets (darkly shaded) at the front
of their input queues are destined for the same upper-right output port. Suppose that
the switch fabric chooses to transfer the packet from the front of the upper-left
queue. In this case, the darkly shaded packet in the lower-left queue must wait. But
not only must this darkly shaded packet wait, so too must the lightly shaded packet
that is queued behind that packet in the lower-left queue, even though there is no
contention for the middle-right output port (the destination for the lightly shaded
packet). This phenomenon is known as head-of-the-line (HOL) blocking in an
Output port contention at time t —
one dark packet can be transferred
Switch
fabric
Light blue packet experiences HOL blocking
Switch
fabric
Key:
destined for upper output
port
Figure 4.11
destined for middle output
port
destined for lower output
port
HOL blocking at an input queued switch
4.4
•
THE INTERNET PROTOCOL (IP)
input-queued switch—a queued packet in an input queue must wait for transfer
through the fabric (even though its output port is free) because it is blocked by
another packet at the head of the line. [Karol 1987] shows that due to HOL blocking, the input queue will grow to unbounded length (informally, this is equivalent to
saying that significant packet loss will occur) under certain assumptions as soon as
the packet arrival rate on the input links reaches only 58 percent of their capacity. A
number of solutions to HOL blocking are discussed in [McKeown 1997b].
4.3.5 The Routing Control Plane
In our discussion thus far and in Figure 4.6, we’ve implicitly assumed that the routing control plane fully resides and executes in a routing processor within the router.
The network-wide routing control plane is thus decentralized—with different pieces
(e.g., of a routing algorithm) executing at different routers and interacting by sending control messages to each other. Indeed, today’s Internet routers and the routing
algorithms we’ll study in Section 4.6 operate in exactly this manner. Additionally,
router and switch vendors bundle their hardware data plane and software control
plane together into closed (but inter-operable) platforms in a vertically integrated
product.
Recently, a number of researchers [Caesar 2005a, Casado 2009, McKeown
2008] have begun exploring new router control plane architectures in which part of
the control plane is implemented in the routers (e.g., local measurement/reporting of
link state, forwarding table installation and maintenance) along with the data plane,
and part of the control plane can be implemented externally to the router (e.g., in a
centralized server, which could perform route calculation). A well-defined API dictates how these two parts interact and communicate with each other. These
researchers argue that separating the software control plane from the hardware data
plane (with a minimal router-resident control plane) can simplify routing by replacing distributed routing calculation with centralized routing calculation, and enable
network innovation by allowing different customized control planes to operate over
fast hardware data planes.
4.4 The Internet Protocol (IP): Forwarding and
Addressing in the Internet
Our discussion of network-layer addressing and forwarding thus far has been
without reference to any specific computer network. In this section, we’ll turn our
attention to how addressing and forwarding are done in the Internet. We’ll see that
Internet addressing and forwarding are important components of the Internet
Protocol (IP). There are two versions of IP in use today. We’ll first examine the
widely deployed IP protocol version 4, which is usually referred to simply as IPv4
331
332
CHAPTER 4
•
THE NETWORK LAYER
Transport layer: TCP, UDP
Routing protocols
• path selection
• RIP, OSPF, BGP
IP protocol
• addressing conventions
• datagram format
• packet handling
conventions
Forwarding
table
Network layer
ICMP protocol
• error reporting
• router “signaling”
Link layer
Physical layer
Figure 4.12
A look inside the Internet’s network layer
[RFC 791]. We’ll examine IP version 6 [RFC 2460; RFC 4291], which has been
proposed to replace IPv4, at the end of this section.
But before beginning our foray into IP, let’s take a step back and consider the
components that make up the Internet’s network layer. As shown in Figure 4.12,
the Internet’s network layer has three major components. The first component is
the IP protocol, the topic of this section. The second major component is the routing component, which determines the path a datagram follows from source to destination. We mentioned earlier that routing protocols compute the forwarding
tables that are used to forward packets through the network. We’ll study the
Internet’s routing protocols in Section 4.6. The final component of the network
layer is a facility to report errors in datagrams and respond to requests for certain
network-layer information. We’ll cover the Internet’s network-layer error- and
information-reporting protocol, the Internet Control Message Protocol (ICMP), in
Section 4.4.3.
4.4.1 Datagram Format
Recall that a network-layer packet is referred to as a datagram. We begin our study
of IP with an overview of the syntax and semantics of the IPv4 datagram. You
might be thinking that nothing could be drier than the syntax and semantics of a
packet’s bits. Nevertheless, the datagram plays a central role in the Internet—every
networking student and professional needs to see it, absorb it, and master it. The
4.4
•
THE INTERNET PROTOCOL (IP)
32 bits
Version
Header
length
Flags 13-bit Fragmentation offset
16-bit Identifier
Time-to-live
Datagram length (bytes)
Type of service
Upper-layer
protocol
Header checksum
32-bit Source IP address
32-bit Destination IP address
Options (if any)
Data
Figure 4.13
IPv4 datagram format
IPv4 datagram format is shown in Figure 4.13. The key fields in the IPv4 datagram
are the following:
• Version number. These 4 bits specify the IP protocol version of the datagram.
By looking at the version number, the router can determine how to interpret
the remainder of the IP datagram. Different versions of IP use different datagram formats. The datagram format for the current version of IP, IPv4, is
shown in Figure 4.13. The datagram format for the new version of IP (IPv6) is
discussed at the end of this section.
• Header length. Because an IPv4 datagram can contain a variable number of
options (which are included in the IPv4 datagram header), these 4 bits are needed
to determine where in the IP datagram the data actually begins. Most IP datagrams do not contain options, so the typical IP datagram has a 20-byte header.
• Type of service. The type of service (TOS) bits were included in the IPv4 header
to allow different types of IP datagrams (for example, datagrams particularly
requiring low delay, high throughput, or reliability) to be distinguished from each
other. For example, it might be useful to distinguish real-time datagrams (such as
those used by an IP telephony application) from non-real-time traffic (for example, FTP). The specific level of service to be provided is a policy issue determined by the router’s administrator. We’ll explore the topic of differentiated
service in Chapter 7.
333
334
CHAPTER 4
•
THE NETWORK LAYER
• Datagram length. This is the total length of the IP datagram (header plus data),
measured in bytes. Since this field is 16 bits long, the theoretical maximum size
of the IP datagram is 65,535 bytes. However, datagrams are rarely larger than
1,500 bytes.
• Identifier, flags, fragmentation offset. These three fields have to do with so-called
IP fragmentation, a topic we will consider in depth shortly. Interestingly, the new
version of IP, IPv6, does not allow for fragmentation at routers.
• Time-to-live. The time-to-live (TTL) field is included to ensure that datagrams
do not circulate forever (due to, for example, a long-lived routing loop) in the
network. This field is decremented by one each time the datagram is processed
by a router. If the TTL field reaches 0, the datagram must be dropped.
• Protocol. This field is used only when an IP datagram reaches its final destination. The value of this field indicates the specific transport-layer protocol to
which the data portion of this IP datagram should be passed. For example, a
value of 6 indicates that the data portion is passed to TCP, while a value of 17
indicates that the data is passed to UDP. For a list of all possible values, see
[IANA Protocol Numbers 2012]. Note that the protocol number in the IP datagram has a role that is analogous to the role of the port number field in the transportlayer segment. The protocol number is the glue that binds the network and transport
layers together, whereas the port number is the glue that binds the transport and
application layers together. We’ll see in Chapter 5 that the link-layer frame also
has a special field that binds the link layer to the network layer.
• Header checksum. The header checksum aids a router in detecting bit errors in a
received IP datagram. The header checksum is computed by treating each 2 bytes
in the header as a number and summing these numbers using 1s complement
arithmetic. As discussed in Section 3.3, the 1s complement of this sum, known
as the Internet checksum, is stored in the checksum field. A router computes the
header checksum for each received IP datagram and detects an error condition if
the checksum carried in the datagram header does not equal the computed checksum. Routers typically discard datagrams for which an error has been detected.
Note that the checksum must be recomputed and stored again at each router, as
the TTL field, and possibly the options field as well, may change. An interesting
discussion of fast algorithms for computing the Internet checksum is [RFC
1071]. A question often asked at this point is, why does TCP/IP perform error
checking at both the transport and network layers? There are several reasons for
this repetition. First, note that only the IP header is checksummed at the IP layer,
while the TCP/UDP checksum is computed over the entire TCP/UDP segment.
Second, TCP/UDP and IP do not necessarily both have to belong to the same protocol stack. TCP can, in principle, run over a different protocol (for example,
ATM) and IP can carry data that will not be passed to TCP/UDP.
• Source and destination IP addresses. When a source creates a datagram, it inserts
its IP address into the source IP address field and inserts the address of the
4.4
•
THE INTERNET PROTOCOL (IP)
ultimate destination into the destination IP address field. Often the source host
determines the destination address via a DNS lookup, as discussed in Chapter 2.
We’ll discuss IP addressing in detail in Section 4.4.2.
• Options. The options fields allow an IP header to be extended. Header options
were meant to be used rarely—hence the decision to save overhead by not
including the information in options fields in every datagram header. However,
the mere existence of options does complicate matters—since datagram headers
can be of variable length, one cannot determine a priori where the data field will
start. Also, since some datagrams may require options processing and others may
not, the amount of time needed to process an IP datagram at a router can vary
greatly. These considerations become particularly important for IP processing in
high-performance routers and hosts. For these reasons and others, IP options
were dropped in the IPv6 header, as discussed in Section 4.4.4.
• Data (payload). Finally, we come to the last and most important field—the raison d’être for the datagram in the first place! In most circumstances, the data
field of the IP datagram contains the transport-layer segment (TCP or UDP) to
be delivered to the destination. However, the data field can carry other types of
data, such as ICMP messages (discussed in Section 4.4.3).
Note that an IP datagram has a total of 20 bytes of header (assuming no options). If
the datagram carries a TCP segment, then each (nonfragmented) datagram carries a
total of 40 bytes of header (20 bytes of IP header plus 20 bytes of TCP header) along
with the application-layer message.
IP Datagram Fragmentation
We’ll see in Chapter 5 that not all link-layer protocols can carry network-layer packets of the same size. Some protocols can carry big datagrams, whereas other protocols can carry only little packets. For example, Ethernet frames can carry up to 1,500
bytes of data, whereas frames for some wide-area links can carry no more than 576
bytes. The maximum amount of data that a link-layer frame can carry is called the
maximum transmission unit (MTU). Because each IP datagram is encapsulated
within the link-layer frame for transport from one router to the next router, the MTU
of the link-layer protocol places a hard limit on the length of an IP datagram. Having
a hard limit on the size of an IP datagram is not much of a problem. What is a problem is that each of the links along the route between sender and destination can use
different link-layer protocols, and each of these protocols can have different MTUs.
To understand the forwarding issue better, imagine that you are a router that
interconnects several links, each running different link-layer protocols with different MTUs. Suppose you receive an IP datagram from one link. You check your forwarding table to determine the outgoing link, and this outgoing link has an MTU
that is smaller than the length of the IP datagram. Time to panic—how are you going
to squeeze this oversized IP datagram into the payload field of the link-layer frame?
335
336
CHAPTER 4
•
THE NETWORK LAYER
The solution is to fragment the data in the IP datagram into two or more smaller IP
datagrams, encapsulate each of these smaller IP datagrams in a separate link-layer
frame; and send these frames over the outgoing link. Each of these smaller datagrams is referred to as a fragment.
Fragments need to be reassembled before they reach the transport layer at the destination. Indeed, both TCP and UDP are expecting to receive complete, unfragmented
segments from the network layer. The designers of IPv4 felt that reassembling datagrams in the routers would introduce significant complication into the protocol and
put a damper on router performance. (If you were a router, would you want to be
reassembling fragments on top of everything else you had to do?) Sticking to the principle of keeping the network core simple, the designers of IPv4 decided to put the job
of datagram reassembly in the end systems rather than in network routers.
When a destination host receives a series of datagrams from the same source, it
needs to determine whether any of these datagrams are fragments of some original,
larger datagram. If some datagrams are fragments, it must further determine when it
has received the last fragment and how the fragments it has received should be
pieced back together to form the original datagram. To allow the destination host to
perform these reassembly tasks, the designers of IP (version 4) put identification,
flag, and fragmentation offset fields in the IP datagram header. When a datagram is
created, the sending host stamps the datagram with an identification number as well
as source and destination addresses. Typically, the sending host increments the identification number for each datagram it sends. When a router needs to fragment a
datagram, each resulting datagram (that is, fragment) is stamped with the source
address, destination address, and identification number of the original datagram.
When the destination receives a series of datagrams from the same sending host, it
can examine the identification numbers of the datagrams to determine which of the
datagrams are actually fragments of the same larger datagram. Because IP is an
unreliable service, one or more of the fragments may never arrive at the destination.
For this reason, in order for the destination host to be absolutely sure it has received
the last fragment of the original datagram, the last fragment has a flag bit set to 0,
whereas all the other fragments have this flag bit set to 1. Also, in order for the destination host to determine whether a fragment is missing (and also to be able to
reassemble the fragments in their proper order), the offset field is used to specify
where the fragment fits within the original IP datagram.
Figure 4.14 illustrates an example. A datagram of 4,000 bytes (20 bytes of IP
header plus 3,980 bytes of IP payload) arrives at a router and must be forwarded to
a link with an MTU of 1,500 bytes. This implies that the 3,980 data bytes in the
original datagram must be allocated to three separate fragments (each of which is
also an IP datagram). Suppose that the original datagram is stamped with an identification number of 777. The characteristics of the three fragments are shown in
Table 4.2. The values in Table 4.2 reflect the requirement that the amount of original payload data in all but the last fragment be a multiple of 8 bytes, and that the offset value be specified in units of 8-byte chunks.
4.4
•
THE INTERNET PROTOCOL (IP)
337
Fragmentation:
In: one large datagram (4,000 bytes)
Out: 3 smaller datagrams
Link MTU: 1,500 bytes
Reassembly:
In: 3 smaller datagrams
Out: one large datagram (4,000 bytes)
Figure 4.14
IP fragmentation and reassembly
At the destination, the payload of the datagram is passed to the transport layer
only after the IP layer has fully reconstructed the original IP datagram. If one or
more of the fragments does not arrive at the destination, the incomplete datagram is
discarded and not passed to the transport layer. But, as we learned in the previous
Fragment
Bytes
ID
Offset
Flag
1st fragment
1,480 bytes in
the data field of
the IP datagram
1,480 bytes
of data
identification ⫽ 777
flag ⫽ 1 (meaning
there is more)
1,020 bytes
(⫽ 3,980–1,480–1,480)
of data
identification ⫽ 777
offset ⫽ 0 (meaning the data
should be inserted beginning
at byte 0)
offset ⫽ 185 (meaning the data
should be inserted beginning at byte
1,480. Note that 185 · 8 ⫽ 1,480)
offset ⫽ 370 (meaning the data
should be inserted beginning at byte
2,960. Note that 370 · 8 ⫽ 2,960)
2nd fragment
3rd fragment
Table 4.2
IP fragments
identification ⫽ 777
flag ⫽ 1 (meaning
there is more)
flag ⫽ 0 (meaning this
is the last fragment)
338
CHAPTER 4
•
THE NETWORK LAYER
chapter, if TCP is being used at the transport layer, then TCP will recover from this
loss by having the source retransmit the data in the original datagram.
We have just learned that IP fragmentation plays an important role in gluing
together the many disparate link-layer technologies. But fragmentation also has its
costs. First, it complicates routers and end systems, which need to be designed to
accommodate datagram fragmentation and reassembly. Second, fragmentation can
be used to create lethal DoS attacks, whereby the attacker sends a series of bizarre
and unexpected fragments. A classic example is the Jolt2 attack, where the attacker
sends a stream of small fragments to the target host, none of which has an offset of
zero. The target can collapse as it attempts to rebuild datagrams out of the degenerate packets. Another class of exploits sends overlapping IP fragments, that is, fragments whose offset values are set so that the fragments do not align properly.
Vulnerable operating systems, not knowing what to do with overlapping fragments,
can crash [Skoudis 2006]. As we’ll see at the end of this section, a new version of
the IP protocol, IPv6, does away with fragmentation altogether, thereby streamlining IP packet processing and making IP less vulnerable to attack.
At this book’s Web site, we provide a Java applet that generates fragments. You
provide the incoming datagram size, the MTU, and the incoming datagram identification. The applet automatically generates the fragments for you. See http://
www.awl.com/kurose-ross.
4.4.2 IPv4 Addressing
We now turn our attention to IPv4 addressing. Although you may be thinking that
addressing must be a straightforward topic, hopefully by the end of this chapter
you’ll be convinced that Internet addressing is not only a juicy, subtle, and interesting topic but also one that is of central importance to the Internet. Excellent treatments of IPv4 addressing are [3Com Addressing 2012] and the first chapter in
[Stewart 1999].
Before discussing IP addressing, however, we’ll need to say a few words about
how hosts and routers are connected into the network. A host typically has only a
single link into the network; when IP in the host wants to send a datagram, it does
so over this link. The boundary between the host and the physical link is called an
interface. Now consider a router and its interfaces. Because a router’s job is to
receive a datagram on one link and forward the datagram on some other link, a
router necessarily has two or more links to which it is connected. The boundary
between the router and any one of its links is also called an interface. A router thus
has multiple interfaces, one for each of its links. Because every host and router is
capable of sending and receiving IP datagrams, IP requires each host and router
interface to have its own IP address. Thus, an IP address is technically associated
with an interface, rather than with the host or router containing that interface.
Each IP address is 32 bits long (equivalently, 4 bytes), and there are thus a total
of 232 possible IP addresses. By approximating 210 by 103, it is easy to see that there
4.4
•
THE INTERNET PROTOCOL (IP)
are about 4 billion possible IP addresses. These addresses are typically written in
so-called dotted-decimal notation, in which each byte of the address is written in
its decimal form and is separated by a period (dot) from other bytes in the address.
For example, consider the IP address 193.32.216.9. The 193 is the decimal equivalent of the first 8 bits of the address; the 32 is the decimal equivalent of the second
8 bits of the address, and so on. Thus, the address 193.32.216.9 in binary notation is
11000001 00100000 11011000 00001001
Each interface on every host and router in the global Internet must have an IP
address that is globally unique (except for interfaces behind NATs, as discussed at
the end of this section). These addresses cannot be chosen in a willy-nilly manner,
however. A portion of an interface’s IP address will be determined by the subnet to
which it is connected.
Figure 4.15 provides an example of IP addressing and interfaces. In this figure,
one router (with three interfaces) is used to interconnect seven hosts. Take a close look
at the IP addresses assigned to the host and router interfaces, as there are several things
to notice. The three hosts in the upper-left portion of Figure 4.15, and the router interface to which they are connected, all have an IP address of the form 223.1.1.xxx. That
is, they all have the same leftmost 24 bits in their IP address. The four interfaces are
also interconnected to each other by a network that contains no routers. This network
223.1.1.1
223.1.1.4
223.1.2.9
223.1.2.1
223.1.3.27
223.1.1.2
223.1.2.2
223.1.1.3
223.1.3.1
Figure 4.15
223.1.3.2
Interface addresses and subnets
339
340
CHAPTER 4
•
THE NETWORK LAYER
could be interconnected by an Ethernet LAN, in which case the interfaces would be
interconnected by an Ethernet switch (as we’ll discuss in Chapter 5), or by a wireless
access point (as we’ll discuss in Chapter 6). We’ll represent this routerless network
connecting these hosts as a cloud for now, and dive into the internals of such networks
in Chapters 5 and 6.
In IP terms, this network interconnecting three host interfaces and one router
interface forms a subnet [RFC 950]. (A subnet is also called an IP network or
simply a network in the Internet literature.) IP addressing assigns an address to this
subnet: 223.1.1.0/24, where the /24 notation, sometimes known as a subnet mask,
indicates that the leftmost 24 bits of the 32-bit quantity define the subnet
address. The subnet 223.1.1.0/24 thus consists of the three host interfaces
(223.1.1.1, 223.1.1.2, and 223.1.1.3) and one router interface (223.1.1.4). Any additional hosts attached to the 223.1.1.0/24 subnet would be required to have an
address of the form 223.1.1.xxx. There are two additional subnets shown in Figure
4.15: the 223.1.2.0/24 network and the 223.1.3.0/24 subnet. Figure 4.16 illustrates
the three IP subnets present in Figure 4.15.
The IP definition of a subnet is not restricted to Ethernet segments that connect
multiple hosts to a router interface. To get some insight here, consider Figure 4.17,
which shows three routers that are interconnected with each other by point-to-point
links. Each router has three interfaces, one for each point-to-point link and one for
the broadcast link that directly connects the router to a pair of hosts. What subnets
are present here? Three subnets, 223.1.1.0/24, 223.1.2.0/24, and 223.1.3.0/24, are
similar to the subnets we encountered in Figure 4.15. But note that there are three
223.1.1.0/23
223.1.2.0/23
223.1.3.0/23
Figure 4.16
Subnet addresses
4.4
•
THE INTERNET PROTOCOL (IP)
additional subnets in this example as well: one subnet, 223.1.9.0/24, for the interfaces
that connect routers R1 and R2; another subnet, 223.1.8.0/24, for the interfaces that
connect routers R2 and R3; and a third subnet, 223.1.7.0/24, for the interfaces that
connect routers R3 and R1. For a general interconnected system of routers and hosts,
we can use the following recipe to define the subnets in the system:
To determine the subnets, detach each interface from its host or router, creating
islands of isolated networks, with interfaces terminating the end points of the
isolated networks. Each of these isolated networks is called a subnet.
If we apply this procedure to the interconnected system in Figure 4.17, we get six
islands or subnets.
From the discussion above, it’s clear that an organization (such as a company
or academic institution) with multiple Ethernet segments and point-to-point links
will have multiple subnets, with all of the devices on a given subnet having the same
subnet address. In principle, the different subnets could have quite different subnet
addresses. In practice, however, their subnet addresses often have much in common.
To understand why, let’s next turn our attention to how addressing is handled in the
global Internet.
223.1.1.1
223.1.1.4
223.1.1.3
223.1.9.2
R1
223.1.7.0
223.1.9.1
R2
223.1.2.6
223.1.2.1
Figure 4.17
223.1.7.1
223.1.8.1
223.1.2.2
223.1.8.0
R3
223.1.3.27
223.1.3.1
Three routers interconnecting six subnets
223.1.3.2
341
342
CHAPTER 4
•
THE NETWORK LAYER
The Internet’s address assignment strategy is known as Classless Interdomain
Routing (CIDR—pronounced cider) [RFC 4632]. CIDR generalizes the notion of
subnet addressing. As with subnet addressing, the 32-bit IP address is divided into
two parts and again has the dotted-decimal form a.b.c.d/x, where x indicates the
number of bits in the first part of the address.
The x most significant bits of an address of the form a.b.c.d/x constitute the
network portion of the IP address, and are often referred to as the prefix (or network prefix) of the address. An organization is typically assigned a block of contiguous addresses, that is, a range of addresses with a common prefix (see the
Principles in Practice sidebar). In this case, the IP addresses of devices within the
organization will share the common prefix. When we cover the Internet’s BGP
PRINCIPLES IN PRACTICE
This example of an ISP that connects eight organizations to the Internet nicely illustrates
how carefully allocated CIDRized addresses facilitate routing. Suppose, as shown in Figure
4.18, that the ISP (which we’ll call Fly-By-Night-ISP) advertises to the outside world that it
should be sent any datagrams whose first 20 address bits match 200.23.16.0/20. The
rest of the world need not know that within the address block 200.23.16.0/20 there are
in fact eight other organizations, each with its own subnets. This ability to use a single prefix to advertise multiple networks is often referred to as address aggregation (also
route aggregation or route summarization).
Address aggregation works extremely well when addresses are allocated in blocks to
ISPs and then from ISPs to client organizations. But what happens when addresses are
not allocated in such a hierarchical manner? What would happen, for example, if Fly-ByNight-ISP acquires ISPs-R-Us and then has Organization 1 connect to the Internet through
its subsidiary ISPs-R-Us? As shown in Figure 4.18, the subsidiary ISPs-R-Us owns the
address block 199.31.0.0/16, but Organization 1’s IP addresses are unfortunately outside of this address block. What should be done here? Certainly, Organization 1 could
renumber all of its routers and hosts to have addresses within the ISPs-R-Us address
block. But this is a costly solution, and Organization 1 might well be reassigned to
another subsidiary in the future. The solution typically adopted is for Organization 1
to keep its IP addresses in 200.23.18.0/23. In this case, as shown in Figure 4.19,
Fly-By-Night-ISP continues to advertise the address block 200.23.16.0/20 and ISPs-R-Us
continues to advertise 199.31.0.0/16. However, ISPs-R-Us now also advertises the block
of addresses for Organization 1, 200.23.18.0/23. When other routers in the larger
Internet see the address blocks 200.23.16.0/20 (from Fly-By-Night-ISP) and
200.23.18.0/23 (from ISPs-R-Us) and want to route to an address in the block
200.23.18.0/23, they will use longest prefix matching (see Section 4.2.2), and route
toward ISPs-R-Us, as it advertises the longest (most specific) address prefix that matches
the destination address.
4.4
•
THE INTERNET PROTOCOL (IP)
Organization 0
200.23.16.0/23
“Send me anything
with addresses
beginning
200.23.16.0/20”
Organization 1
200.23.18.0/23
Fly-By-Night-ISP
Organization 2
200.23.20.0/23
Internet
Organization 7
“Send me anything
with addresses
beginning
199.31.0.0/16”
200.23.30.0/23
ISPs-R-Us
Figure 4.18
Hierarchical addressing and route aggregation
Organization 0
200.23.16.0/23
“Send me anything
with addresses
beginning
200.23.16.0/20”
Organization 2
200.23.20.0/23
Fly-By-Night-ISP
Organization 7
200.23.30.0/23
“Send me anything
with addresses
beginning
199.31.0.0/16 or
200.23.18.0/23”
Organization 1
200.23.18.0/23
Internet
ISPs-R-Us
Figure 4.19
ISPs-R-Us has a more specific route to Organization 1
343
344
CHAPTER 4
•
THE NETWORK LAYER
routing protocol in Section 4.6, we’ll see that only these x leading prefix bits are
considered by routers outside the organization’s network. That is, when a router
outside the organization forwards a datagram whose destination address is inside
the organization, only the leading x bits of the address need be considered. This
considerably reduces the size of the forwarding table in these routers, since a single entry of the form a.b.c.d/x will be sufficient to forward packets to any destination within the organization.
The remaining 32-x bits of an address can be thought of as distinguishing
among the devices within the organization, all of which have the same network prefix. These are the bits that will be considered when forwarding packets at routers
within the organization. These lower-order bits may (or may not) have an additional
subnetting structure, such as that discussed above. For example, suppose the first 21
bits of the CIDRized address a.b.c.d/21 specify the organization’s network prefix
and are common to the IP addresses of all devices in that organization. The remaining 11 bits then identify the specific hosts in the organization. The organization’s
internal structure might be such that these 11 rightmost bits are used for subnetting
within the organization, as discussed above. For example, a.b.c.d/24 might refer to a
specific subnet within the organization.
Before CIDR was adopted, the network portions of an IP address were constrained to be 8, 16, or 24 bits in length, an addressing scheme known as classful
addressing, since subnets with 8-, 16-, and 24-bit subnet addresses were known as
class A, B, and C networks, respectively. The requirement that the subnet portion of
an IP address be exactly 1, 2, or 3 bytes long turned out to be problematic for supporting the rapidly growing number of organizations with small and medium-sized
subnets. A class C (/24) subnet could accommodate only up to 28 – 2 = 254 hosts
(two of the 28 = 256 addresses are reserved for special use)—too small for many
organizations. However, a class B (/16) subnet, which supports up to 65,634 hosts,
was too large. Under classful addressing, an organization with, say, 2,000 hosts was
typically allocated a class B (/16) subnet address. This led to a rapid depletion of the
class B address space and poor utilization of the assigned address space. For example, the organization that used a class B address for its 2,000 hosts was allocated
enough of the address space for up to 65,534 interfaces—leaving more than 63,000
addresses that could not be used by other organizations.
We would be remiss if we did not mention yet another type of IP address, the IP
broadcast address 255.255.255.255. When a host sends a datagram with destination
address 255.255.255.255, the message is delivered to all hosts on the same subnet.
Routers optionally forward the message into neighboring subnets as well (although
they usually don’t).
Having now studied IP addressing in detail, we need to know how hosts and
subnets get their addresses in the first place. Let’s begin by looking at how an
organization gets a block of addresses for its devices, and then look at how a device
(such as a host) is assigned an address from within the organization’s block of
addresses.
4.4
•
THE INTERNET PROTOCOL (IP)
Obtaining a Block of Addresses
In order to obtain a block of IP addresses for use within an organization’s subnet, a
network administrator might first contact its ISP, which would provide addresses
from a larger block of addresses that had already been allocated to the ISP. For
example, the ISP may itself have been allocated the address block 200.23.16.0/20.
The ISP, in turn, could divide its address block into eight equal-sized contiguous
address blocks and give one of these address blocks out to each of up to eight organizations that are supported by this ISP, as shown below. (We have underlined the
subnet part of these addresses for your convenience.)
ISP’s block
200.23.16.0/20
11001000 00010111 00010000 00000000
Organization 0
200.23.16.0/23
11001000 00010111 00010000 00000000
Organization 1
200.23.18.0/23
11001000 00010111 00010010 00000000
Organization 2
200.23.20.0/23
11001000 00010111 00010100 00000000
...
...
Organization 7
...
200.23.30.0/23
11001000 00010111 00011110 00000000
While obtaining a set of addresses from an ISP is one way to get a block of
addresses, it is not the only way. Clearly, there must also be a way for the ISP itself
to get a block of addresses. Is there a global authority that has ultimate responsibility
for managing the IP address space and allocating address blocks to ISPs and other
organizations? Indeed there is! IP addresses are managed under the authority of the
Internet Corporation for Assigned Names and Numbers (ICANN) [ICANN 2012],
based on guidelines set forth in [RFC 2050]. The role of the nonprofit ICANN organization [NTIA 1998] is not only to allocate IP addresses, but also to manage the DNS
root servers. It also has the very contentious job of assigning domain names and
resolving domain name disputes. The ICANN allocates addresses to regional Internet registries (for example, ARIN, RIPE, APNIC, and LACNIC, which together
form the Address Supporting Organization of ICANN [ASO-ICANN 2012]), and
handle the allocation/management of addresses within their regions.
Obtaining a Host Address: the Dynamic Host Configuration Protocol
Once an organization has obtained a block of addresses, it can assign individual IP
addresses to the host and router interfaces in its organization. A system administrator will typically manually configure the IP addresses into the router (often
remotely, with a network management tool). Host addresses can also be configured
manually, but more often this task is now done using the Dynamic Host Configuration Protocol (DHCP) [RFC 2131]. DHCP allows a host to obtain (be allocated)
an IP address automatically. A network administrator can configure DHCP so that a
345
346
CHAPTER 4
•
THE NETWORK LAYER
given host receives the same IP address each time it connects to the network, or a
host may be assigned a temporary IP address that will be different each time the
host connects to the network. In addition to host IP address assignment, DHCP also
allows a host to learn additional information, such as its subnet mask, the address of
its first-hop router (often called the default gateway), and the address of its local
DNS server.
Because of DHCP’s ability to automate the network-related aspects of connecting a host into a network, it is often referred to as a plug-and-play protocol. This
capability makes it very attractive to the network administrator who would otherwise have to perform these tasks manually! DHCP is also enjoying widespread use
in residential Internet access networks and in wireless LANs, where hosts join and
leave the network frequently. Consider, for example, the student who carries a laptop from a dormitory room to a library to a classroom. It is likely that in each location, the student will be connecting into a new subnet and hence will need a new IP
address at each location. DHCP is ideally suited to this situation, as there are many
users coming and going, and addresses are needed for only a limited amount of time.
DHCP is similarly useful in residential ISP access networks. Consider, for example,
a residential ISP that has 2,000 customers, but no more than 400 customers are ever
online at the same time. In this case, rather than needing a block of 2,048 addresses,
a DHCP server that assigns addresses dynamically needs only a block of 512
addresses (for example, a block of the form a.b.c.d/23). As the hosts join and leave,
the DHCP server needs to update its list of available IP addresses. Each time a host
joins, the DHCP server allocates an arbitrary address from its current pool of available addresses; each time a host leaves, its address is returned to the pool.
DHCP is a client-server protocol. A client is typically a newly arriving host
wanting to obtain network configuration information, including an IP address for
itself. In the simplest case, each subnet (in the addressing sense of Figure 4.17) will
have a DHCP server. If no server is present on the subnet, a DHCP relay agent (typically a router) that knows the address of a DHCP server for that network is needed.
Figure 4.20 shows a DHCP server attached to subnet 223.1.2/24, with the router
serving as the relay agent for arriving clients attached to subnets 223.1.1/24 and
223.1.3/24. In our discussion below, we’ll assume that a DHCP server is available
on the subnet.
For a newly arriving host, the DHCP protocol is a four-step process, as shown
in Figure 4.21 for the network setting shown in Figure 4.20. In this figure, yiaddr
(as in “your Internet address”) indicates the address being allocated to the newly
arriving client. The four steps are:
• DHCP server discovery. The first task of a newly arriving host is to find a DHCP
server with which to interact. This is done using a DHCP discover message,
which a client sends within a UDP packet to port 67. The UDP packet is encapsulated in an IP datagram. But to whom should this datagram be sent? The host
doesn’t even know the IP address of the network to which it is attaching, much
4.4
•
THE INTERNET PROTOCOL (IP)
DHCP
server
223.1.2.5
223.1.1.1
223.1.1.4
223.1.2.9
223.1.2.1
223.1.3.27
223.1.1.2
Arriving
DHCP
client
223.1.2.2
223.1.1.3
223.1.3.1
Figure 4.20
223.1.3.2
DHCP client-server scenario
less the address of a DHCP server for this network. Given this, the DHCP client
creates an IP datagram containing its DHCP discover message along with the
broadcast destination IP address of 255.255.255.255 and a “this host” source IP
address of 0.0.0.0. The DHCP client passes the IP datagram to the link layer,
which then broadcasts this frame to all nodes attached to the subnet (we will
cover the details of link-layer broadcasting in Section 5.4).
• DHCP server offer(s). A DHCP server receiving a DHCP discover message
responds to the client with a DHCP offer message that is broadcast to all nodes
on the subnet, again using the IP broadcast address of 255.255.255.255. (You
might want to think about why this server reply must also be broadcast). Since
several DHCP servers can be present on the subnet, the client may find itself in
the enviable position of being able to choose from among several offers. Each
server offer message contains the transaction ID of the received discover message, the proposed IP address for the client, the network mask, and an IP address
lease time—the amount of time for which the IP address will be valid. It is common for the server to set the lease time to several hours or days [Droms 2002].
347
348
CHAPTER 4
•
THE NETWORK LAYER
DHCP server:
223.1.2.5
Arriving client
DHCP discover
src: 0.0.0.0, 68
dest: 255.255.255.255,67
DHCPDISCOVER
yiaddr: 0.0.0.0
transaction ID: 654
DHCP request
DHCP offer
src: 223.1.2.5, 67
dest: 255.255.255.255,68
DHCPOFFER
yiaddrr: 223.1.2.4
transaction ID: 654
DHCP server ID: 223.1.2.5
Lifetime: 3600 secs
src: 0.0.0.0, 68
dest: 255.255.255.255, 67
DHCPREQUEST
yiaddrr: 223.1.2.4
transaction ID: 655
DHCP server ID: 223.1.2.5
Lifetime: 3600 secs
DHCP ACK
src: 223.1.2.5, 67
dest: 255.255.255.255,68
DHCPACK
yiaddrr: 223.1.2.4
transaction ID: 655
DHCP server ID: 223.1.2.5
Lifetime: 3600 secs
Time
Figure 4.21
Time
DHCP client-server interaction
• DHCP request. The newly arriving client will choose from among one or more
server offers and respond to its selected offer with a DHCP request message,
echoing back the configuration parameters.
• DHCP ACK. The server responds to the DHCP request message with a DHCP
ACK message, confirming the requested parameters.
Once the client receives the DHCP ACK, the interaction is complete and the
client can use the DHCP-allocated IP address for the lease duration. Since a client
4.4
•
THE INTERNET PROTOCOL (IP)
may want to use its address beyond the lease’s expiration, DHCP also provides a
mechanism that allows a client to renew its lease on an IP address.
The value of DHCP’s plug-and-play capability is clear, considering the fact that
the alternative is to manually configure a host’s IP address. Consider the student
who moves from classroom to library to dorm room with a laptop, joins a new subnet, and thus obtains a new IP address at each location. It is unimaginable that a system administrator would have to reconfigure laptops at each location, and few
students (except those taking a computer networking class!) would have the expertise to configure their laptops manually. From a mobility aspect, however, DHCP
does have shortcomings. Since a new IP address is obtained from DHCP each time
a node connects to a new subnet, a TCP connection to a remote application cannot
be maintained as a mobile node moves between subnets. In Chapter 6, we will
examine mobile IP—a recent extension to the IP infrastructure that allows a mobile
node to use a single permanent address as it moves between subnets. Additional
details about DHCP can be found in [Droms 2002] and [dhc 2012]. An open source
reference implementation of DHCP is available from the Internet Systems Consortium [ISC 2012].
Network Address Translation (NAT)
Given our discussion about Internet addresses and the IPv4 datagram format, we’re
now well aware that every IP-capable device needs an IP address. With the proliferation of small office, home office (SOHO) subnets, this would seem to imply that
whenever a SOHO wants to install a LAN to connect multiple machines, a range of
addresses would need to be allocated by the ISP to cover all of the SOHO’s
machines. If the subnet grew bigger (for example, the kids at home have not only
their own computers, but have smartphones and networked Game Boys as well), a
larger block of addresses would have to be allocated. But what if the ISP had already
allocated the contiguous portions of the SOHO network’s current address range?
And what typical homeowner wants (or should need) to know how to manage IP
addresses in the first place? Fortunately, there is a simpler approach to address allocation that has found increasingly widespread use in such scenarios: network
address translation (NAT) [RFC 2663; RFC 3022; Zhang 2007].
Figure 4.22 shows the operation of a NAT-enabled router. The NAT-enabled
router, residing in the home, has an interface that is part of the home network on the
right of Figure 4.22. Addressing within the home network is exactly as we have seen
above—all four interfaces in the home network have the same subnet address of
10.0.0/24. The address space 10.0.0.0/8 is one of three portions of the IP address
space that is reserved in [RFC 1918] for a private network or a realm with private
addresses, such as the home network in Figure 4.22. A realm with private addresses
refers to a network whose addresses only have meaning to devices within that
network. To see why this is important, consider the fact that there are hundreds of
349
350
CHAPTER 4
•
THE NETWORK LAYER
NAT translation table
WAN side
138.76.29.7, 5001
...
LAN side
10.0.0.1, 3345
...
S = 10.0.0.1, 3345
D = 128.119.40.186, 80
10.0.0.1
1
2
S = 138.76.29.7, 5001
D = 128.119.40.186, 80
10.0.0.4
138.76.29.7
S = 128.119.40.186, 80
D = 138.76.29.7, 5001
Figure 4.22
4
S = 128.119.40.186, 80
D = 10.0.0.1, 3345
3
10.0.0.2
10.0.0.3
Network address translation
thousands of home networks, many using the same address space, 10.0.0.0/24.
Devices within a given home network can send packets to each other using
10.0.0.0/24 addressing. However, packets forwarded beyond the home network into
the larger global Internet clearly cannot use these addresses (as either a source or a
destination address) because there are hundreds of thousands of networks using this
block of addresses. That is, the 10.0.0.0/24 addresses can only have meaning within
the given home network. But if private addresses only have meaning within a given
network, how is addressing handled when packets are sent to or received from the
global Internet, where addresses are necessarily unique? The answer lies in understanding NAT.
The NAT-enabled router does not look like a router to the outside world. Instead
the NAT router behaves to the outside world as a single device with a single IP
address. In Figure 4.22, all traffic leaving the home router for the larger Internet has
a source IP address of 138.76.29.7, and all traffic entering the home router must
have a destination address of 138.76.29.7. In essence, the NAT-enabled router is hiding the details of the home network from the outside world. (As an aside, you might
wonder where the home network computers get their addresses and where the router
gets its single IP address. Often, the answer is the same—DHCP! The router gets its
address from the ISP’s DHCP server, and the router runs a DHCP server to provide
addresses to computers within the NAT-DHCP-router-controlled home network’s
address space.)
4.4
•
THE INTERNET PROTOCOL (IP)
If all datagrams arriving at the NAT router from the WAN have the same destination IP address (specifically, that of the WAN-side interface of the NAT router),
then how does the router know the internal host to which it should forward a given
datagram? The trick is to use a NAT translation table at the NAT router, and to
include port numbers as well as IP addresses in the table entries.
Consider the example in Figure 4.22. Suppose a user sitting in a home network
behind host 10.0.0.1 requests a Web page on some Web server (port 80) with IP
address 128.119.40.186. The host 10.0.0.1 assigns the (arbitrary) source port number 3345 and sends the datagram into the LAN. The NAT router receives the datagram, generates a new source port number 5001 for the datagram, replaces the
source IP address with its WAN-side IP address 138.76.29.7, and replaces the original source port number 3345 with the new source port number 5001. When generating a new source port number, the NAT router can select any source port number
that is not currently in the NAT translation table. (Note that because a port number
field is 16 bits long, the NAT protocol can support over 60,000 simultaneous connections with a single WAN-side IP address for the router!) NAT in the router also
adds an entry to its NAT translation table. The Web server, blissfully unaware that
the arriving datagram containing the HTTP request has been manipulated by the
NAT router, responds with a datagram whose destination address is the IP address
of the NAT router, and whose destination port number is 5001. When this datagram
arrives at the NAT router, the router indexes the NAT translation table using the destination IP address and destination port number to obtain the appropriate IP address
(10.0.0.1) and destination port number (3345) for the browser in the home network.
The router then rewrites the datagram’s destination address and destination port
number, and forwards the datagram into the home network.
NAT has enjoyed widespread deployment in recent years. But we should
mention that many purists in the IETF community loudly object to NAT. First,
they argue, port numbers are meant to be used for addressing processes, not for
addressing hosts. (This violation can indeed cause problems for servers running
on the home network, since, as we have seen in Chapter 2, server processes wait
for incoming requests at well-known port numbers.) Second, they argue, routers
are supposed to process packets only up to layer 3. Third, they argue, the NAT
protocol violates the so-called end-to-end argument; that is, hosts should be talking directly with each other, without interfering nodes modifying IP addresses and
port numbers. And fourth, they argue, we should use IPv6 (see Section 4.4.4) to
solve the shortage of IP addresses, rather than recklessly patching up the problem
with a stopgap solution like NAT. But like it or not, NAT has become an important
component of the Internet.
Yet another major problem with NAT is that it interferes with P2P applications,
including P2P file-sharing applications and P2P Voice-over-IP applications. Recall
from Chapter 2 that in a P2P application, any participating Peer A should be able to
initiate a TCP connection to any other participating Peer B. The essence of the
problem is that if Peer B is behind a NAT, it cannot act as a server and accept TCP
351
352
CHAPTER 4
•
THE NETWORK LAYER
connections. As we’ll see in the homework problems, this NAT problem can be circumvented if Peer A is not behind a NAT. In this case, Peer A can first contact Peer
B through an intermediate Peer C, which is not behind a NAT and to which B has
established an ongoing TCP connection. Peer A can then ask Peer B, via Peer C, to
initiate a TCP connection directly back to Peer A. Once the direct P2P TCP connection is established between Peers A and B, the two peers can exchange messages or
files. This hack, called connection reversal, is actually used by many P2P applications for NAT traversal. If both Peer A and Peer B are behind their own NATs, the
situation is a bit trickier but can be handled using application relays, as we saw with
Skype relays in Chapter 2.
UPnP
NAT traversal is increasingly provided by Universal Plug and Play (UPnP), which is
a protocol that allows a host to discover and configure a nearby NAT [UPnP Forum
2012]. UPnP requires that both the host and the NAT be UPnP compatible. With
UPnP, an application running in a host can request a NAT mapping between its
( private IP address, private port number) and the ( public IP address, public port
number) for some requested public port number. If the NAT accepts the request and
creates the mapping, then nodes from the outside can initiate TCP connections to
( public IP address, public port number). Furthermore, UPnP lets the application
know the value of ( public IP address, public port number), so that the application
can advertise it to the outside world.
As an example, suppose your host, behind a UPnP-enabled NAT, has private
address 10.0.0.1 and is running BitTorrent on port 3345. Also suppose that the
public IP address of the NAT is 138.76.29.7. Your BitTorrent application naturally
wants to be able to accept connections from other hosts, so that it can trade chunks
with them. To this end, the BitTorrent application in your host asks the NAT to create a “hole” that maps (10.0.0.1, 3345) to (138.76.29.7, 5001). (The public port
number 5001 is chosen by the application.) The BitTorrent application in your host
could also advertise to its tracker that it is available at (138.76.29.7, 5001). In this
manner, an external host running BitTorrent can contact the tracker and learn that
your BitTorrent application is running at (138.76.29.7, 5001). The external host
can send a TCP SYN packet to (138.76.29.7, 5001). When the NAT receives the
SYN packet, it will change the destination IP address and port number in the
packet to (10.0.0.1, 3345) and forward the packet through the NAT.
In summary, UPnP allows external hosts to initiate communication sessions
to NATed hosts, using either TCP or UDP. NATs have long been a nemesis
for P2P applications; UPnP, providing an effective and robust NAT traversal
solution, may be their savior. Our discussion of NAT and UPnP here has been
necessarily brief. For more detailed discussions of NAT see [Huston 2004, Cisco
NAT 2012].
4.4
•
THE INTERNET PROTOCOL (IP)
4.4.3 Internet Control Message Protocol (ICMP)
Recall that the network layer of the Internet has three main components: the IP protocol, discussed in the previous section; the Internet routing protocols (including
RIP, OSPF, and BGP), which are covered in Section 4.6; and ICMP, which is the
subject of this section.
ICMP, specified in [RFC 792], is used by hosts and routers to communicate network-layer information to each other. The most typical use of ICMP is for error
reporting. For example, when running a Telnet, FTP, or HTTP session, you may
have encountered an error message such as “Destination network unreachable.” This
message had its origins in ICMP. At some point, an IP router was unable to find a
path to the host specified in your Telnet, FTP, or HTTP application. That router created and sent a type-3 ICMP message to your host indicating the error.
ICMP is often considered part of IP but architecturally it lies just above IP, as
ICMP messages are carried inside IP datagrams. That is, ICMP messages are carried
as IP payload, just as TCP or UDP segments are carried as IP payload. Similarly,
when a host receives an IP datagram with ICMP specified as the upper-layer protocol, it demultiplexes the datagram’s contents to ICMP, just as it would demultiplex a
datagram’s content to TCP or UDP.
ICMP messages have a type and a code field, and contain the header and the
first 8 bytes of the IP datagram that caused the ICMP message to be generated in the
first place (so that the sender can determine the datagram that caused the error).
Selected ICMP message types are shown in Figure 4.23. Note that ICMP messages
are used not only for signaling error conditions.
The well-known ping program sends an ICMP type 8 code 0 message to the
specified host. The destination host, seeing the echo request, sends back a type 0
code 0 ICMP echo reply. Most TCP/IP implementations support the ping server
directly in the operating system; that is, the server is not a process. Chapter 11 of
[Stevens 1990] provides the source code for the ping client program. Note that the
client program needs to be able to instruct the operating system to generate an ICMP
message of type 8 code 0.
Another interesting ICMP message is the source quench message. This message
is seldom used in practice. Its original purpose was to perform congestion control—
to allow a congested router to send an ICMP source quench message to a host to
force that host to reduce its transmission rate. We have seen in Chapter 3 that TCP
has its own congestion-control mechanism that operates at the transport layer, without the use of network-layer feedback such as the ICMP source quench message.
In Chapter 1 we introduced the Traceroute program, which allows us to trace a
route from a host to any other host in the world. Interestingly, Traceroute is implemented with ICMP messages. To determine the names and addresses of the routers
between source and destination, Traceroute in the source sends a series of ordinary
IP datagrams to the destination. Each of these datagrams carries a UDP segment
with an unlikely UDP port number. The first of these datagrams has a TTL of 1, the
353
354
CHAPTER 4
•
THE NETWORK LAYER
ICMP Type
Code
0
0
echo reply (to ping)
3
0
destination network unreachable
3
1
destination host unreachable
3
2
destination protocol unreachable
3
3
destination port unreachable
3
6
destination network unknown
3
7
destination host unknown
4
0
source quench (congestion control)
8
0
echo request
9
0
router advertisement
10
0
router discovery
11
0
TTL expired
12
0
IP header bad
Figure 4.23
Description
ICMP message types
second of 2, the third of 3, and so on. The source also starts timers for each of the
datagrams. When the nth datagram arrives at the nth router, the nth router observes
that the TTL of the datagram has just expired. According to the rules of the IP protocol, the router discards the datagram and sends an ICMP warning message to the
source (type 11 code 0). This warning message includes the name of the router and
its IP address. When this ICMP message arrives back at the source, the source
obtains the round-trip time from the timer and the name and IP address of the nth
router from the ICMP message.
How does a Traceroute source know when to stop sending UDP segments?
Recall that the source increments the TTL field for each datagram it sends. Thus,
one of the datagrams will eventually make it all the way to the destination host.
Because this datagram contains a UDP segment with an unlikely port number, the
destination host sends a port unreachable ICMP message (type 3 code 3) back to the
source. When the source host receives this particular ICMP message, it knows it
does not need to send additional probe packets. (The standard Traceroute program
actually sends sets of three packets with the same TTL; thus the Traceroute output
provides three results for each TTL.)
4.4
•
THE INTERNET PROTOCOL (IP)
FOCUS ON SECURITY
INSPECTING DATAGRAMS: FIREWALLS AND INTRUSION DETECTION
SYSTEMS
Suppose you are assigned the task of administering a home, departmental, university, or
corporate network. Attackers, knowing the IP address range of your network, can easily
send IP datagrams to addresses in your range. These datagrams can do all kinds of
devious things, including mapping your network with ping sweeps and port scans,
crashing vulnerable hosts with malformed packets, flooding servers with a deluge of
ICMP packets, and infecting hosts by including malware in the packets. As the network
administrator, what are you going to do about all those bad guys out there, each capable of sending malicious packets into your network? Two popular defense mechanisms
to malicious packet attacks are firewalls and intrusion detection systems (IDSs).
As a network administrator, you may first try installing a firewall between your
network and the Internet. (Most access routers today have firewall capability.)
Firewalls inspect the datagram and segment header fields, denying suspicious datagrams entry into the internal network. For example, a firewall may be configured to
block all ICMP echo request packets, thereby preventing an attacker from doing a
traditional ping sweep across your IP address range. Firewalls can also block packets based on source and destination IP addresses and port numbers. Additionally,
firewalls can be configured to track TCP connections, granting entry only to datagrams that belong to approved connections.
Additional protection can be provided with an IDS. An IDS, typically situated at the
network boundary, performs “deep packet inspection,” examining not only header
fields but also the payloads in the datagram (including application-layer data). An IDS
has a database of packet signatures that are known to be part of attacks. This database is automatically updated as new attacks are discovered. As packets pass through
the IDS, the IDS attempts to match header fields and payloads to the signatures in its
signature database. If such a match is found, an alert is created. An intrusion prevention system (IPS) is similar to an IDS, except that it actually blocks packets in addition to
creating alerts. In Chapter 8, we’ll explore firewalls and IDSs in more detail.
Can firewalls and IDSs fully shield your network from all attacks? The answer is
clearly no, as attackers continually find new attacks for which signatures are not yet
available. But firewalls and traditional signature-based IDSs are useful in protecting
your network from known attacks.
In this manner, the source host learns the number and the identities of routers
that lie between it and the destination host and the round-trip time between the two
hosts. Note that the Traceroute client program must be able to instruct the operating
system to generate UDP datagrams with specific TTL values and must also be able to
be notified by its operating system when ICMP messages arrive. Now that you understand how Traceroute works, you may want to go back and play with it some more.
355
356
CHAPTER 4
•
THE NETWORK LAYER
4.4.4 IPv6
In the early 1990s, the Internet Engineering Task Force began an effort to develop a
successor to the IPv4 protocol. A prime motivation for this effort was the realization
that the 32-bit IP address space was beginning to be used up, with new subnets and
IP nodes being attached to the Internet (and being allocated unique IP addresses) at
a breathtaking rate. To respond to this need for a large IP address space, a new IP
protocol, IPv6, was developed. The designers of IPv6 also took this opportunity to
tweak and augment other aspects of IPv4, based on the accumulated operational
experience with IPv4.
The point in time when IPv4 addresses would be completely allocated (and
hence no new networks could attach to the Internet) was the subject of considerable
debate. The estimates of the two leaders of the IETF’s Address Lifetime Expectations working group were that addresses would become exhausted in 2008 and 2018,
respectively [Solensky 1996]. In February 2011, IANA allocated out the last remaining pool of unassigned IPv4 addresses to a regional registry. While these registries
still have available IPv4 addresses within their pool, once these addresses are
exhausted, there are no more available address blocks that can be allocated from a
central pool [Huston 2011a]. Although the mid-1990s estimates of IPv4 address
depletion suggested that a considerable amount of time might be left until the IPv4
address space was exhausted, it was realized that considerable time would be needed
to deploy a new technology on such an extensive scale, and so the Next Generation
IP (IPng) effort [Bradner 1996; RFC 1752] was begun. The result of this effort was
the specification of IP version 6 (IPv6) [RFC 2460] which we’ll discuss below. (An
often-asked question is what happened to IPv5? It was initially envisioned that the
ST-2 protocol would become IPv5, but ST-2 was later dropped.) Excellent sources of
information about IPv6 are [Huitema 1998, IPv6 2012].
IPv6 Datagram Format
The format of the IPv6 datagram is shown in Figure 4.24. The most important
changes introduced in IPv6 are evident in the datagram format:
• Expanded addressing capabilities. IPv6 increases the size of the IP address
from 32 to 128 bits. This ensures that the world won’t run out of IP addresses.
Now, every grain of sand on the planet can be IP-addressable. In addition to
unicast and multicast addresses, IPv6 has introduced a new type of address,
called an anycast address, which allows a datagram to be delivered to any
one of a group of hosts. (This feature could be used, for example, to send an
HTTP GET to the nearest of a number of mirror sites that contain a given
document.)
• A streamlined 40-byte header. As discussed below, a number of IPv4 fields have
been dropped or made optional. The resulting 40-byte fixed-length header allows
4.4
•
THE INTERNET PROTOCOL (IP)
32 bits
Version
Traffic class
Flow label
Payload length
Next hdr
Hop limit
Source address
(128 bits)
Destination address
(128 bits)
Data
Figure 4.24
IPv6 datagram format
for faster processing of the IP datagram. A new encoding of options allows for
more flexible options processing.
• Flow labeling and priority. IPv6 has an elusive definition of a flow. RFC 1752
and RFC 2460 state that this allows “labeling of packets belonging to particular
flows for which the sender requests special handling, such as a nondefault quality
of service or real-time service.” For example, audio and video transmission might
likely be treated as a flow. On the other hand, the more traditional applications,
such as file transfer and e-mail, might not be treated as flows. It is possible that the
traffic carried by a high-priority user (for example, someone paying for better service for their traffic) might also be treated as a flow. What is clear, however, is that
the designers of IPv6 foresee the eventual need to be able to differentiate among
the flows, even if the exact meaning of a flow has not yet been determined. The
IPv6 header also has an 8-bit traffic class field. This field, like the TOS field in
IPv4, can be used to give priority to certain datagrams within a flow, or it can be
used to give priority to datagrams from certain applications (for example, ICMP)
over datagrams from other applications (for example, network news).
As noted above, a comparison of Figure 4.24 with Figure 4.13 reveals the simpler, more streamlined structure of the IPv6 datagram. The following fields are
defined in IPv6:
• Version. This 4-bit field identifies the IP version number. Not surprisingly, IPv6
carries a value of 6 in this field. Note that putting a 4 in this field does not create
a valid IPv4 datagram. (If it did, life would be a lot simpler—see the discussion
below regarding the transition from IPv4 to IPv6.)
357
358
CHAPTER 4
•
THE NETWORK LAYER
• Traffic class. This 8-bit field is similar in spirit to the TOS field we saw in IPv4.
• Flow label. As discussed above, this 20-bit field is used to identify a flow of
datagrams.
• Payload length. This 16-bit value is treated as an unsigned integer giving the
number of bytes in the IPv6 datagram following the fixed-length, 40-byte datagram header.
• Next header. This field identifies the protocol to which the contents (data field)
of this datagram will be delivered (for example, to TCP or UDP). The field uses
the same values as the protocol field in the IPv4 header.
• Hop limit. The contents of this field are decremented by one by each router that
forwards the datagram. If the hop limit count reaches zero, the datagram is
discarded.
• Source and destination addresses. The various formats of the IPv6 128-bit
address are described in RFC 4291.
• Data. This is the payload portion of the IPv6 datagram. When the datagram
reaches its destination, the payload will be removed from the IP datagram and
passed on to the protocol specified in the next header field.
The discussion above identified the purpose of the fields that are included in the
IPv6 datagram. Comparing the IPv6 datagram format in Figure 4.24 with the IPv4
datagram format that we saw in Figure 4.13, we notice that several fields appearing
in the IPv4 datagram are no longer present in the IPv6 datagram:
• Fragmentation/Reassembly. IPv6 does not allow for fragmentation and reassembly at intermediate routers; these operations can be performed only by the source
and destination. If an IPv6 datagram received by a router is too large to be forwarded over the outgoing link, the router simply drops the datagram and sends a
“Packet Too Big” ICMP error message (see below) back to the sender. The
sender can then resend the data, using a smaller IP datagram size. Fragmentation
and reassembly is a time-consuming operation; removing this functionality from
the routers and placing it squarely in the end systems considerably speeds up IP
forwarding within the network.
• Header checksum. Because the transport-layer (for example, TCP and UDP) and
link-layer (for example, Ethernet) protocols in the Internet layers perform checksumming, the designers of IP probably felt that this functionality was sufficiently
redundant in the network layer that it could be removed. Once again, fast processing of IP packets was a central concern. Recall from our discussion of IPv4
in Section 4.4.1 that since the IPv4 header contains a TTL field (similar to the
hop limit field in IPv6), the IPv4 header checksum needed to be recomputed at
every router. As with fragmentation and reassembly, this too was a costly operation in IPv4.
4.4
•
THE INTERNET PROTOCOL (IP)
• Options. An options field is no longer a part of the standard IP header. However, it has not gone away. Instead, the options field is one of the possible next
headers pointed to from within the IPv6 header. That is, just as TCP or UDP
protocol headers can be the next header within an IP packet, so too can an
options field. The removal of the options field results in a fixed-length, 40byte IP header.
Recall from our discussion in Section 4.4.3 that the ICMP protocol is used by IP
nodes to report error conditions and provide limited information (for example, the
echo reply to a ping message) to an end system. A new version of ICMP has been
defined for IPv6 in RFC 4443. In addition to reorganizing the existing ICMP type
and code definitions, ICMPv6 also added new types and codes required by the new
IPv6 functionality. These include the “Packet Too Big” type, and an “unrecognized
IPv6 options” error code. In addition, ICMPv6 subsumes the functionality of the
Internet Group Management Protocol (IGMP) that we’ll study in Section 4.7. IGMP,
which is used to manage a host’s joining and leaving of multicast groups, was previously a separate protocol from ICMP in IPv4.
Transitioning from IPv4 to IPv6
Now that we have seen the technical details of IPv6, let us consider a very practical
matter: How will the public Internet, which is based on IPv4, be transitioned to
IPv6? The problem is that while new IPv6-capable systems can be made backwardcompatible, that is, can send, route, and receive IPv4 datagrams, already deployed
IPv4-capable systems are not capable of handling IPv6 datagrams. Several options
are possible [Huston 2011b].
One option would be to declare a flag day—a given time and date when all
Internet machines would be turned off and upgraded from IPv4 to IPv6. The last
major technology transition (from using NCP to using TCP for reliable transport
service) occurred almost 25 years ago. Even back then [RFC 801], when the Internet was tiny and still being administered by a small number of “wizards,” it was
realized that such a flag day was not possible. A flag day involving hundreds of millions of machines and millions of network administrators and users is even more
unthinkable today. RFC 4213 describes two approaches (which can be used either
alone or together) for gradually integrating IPv6 hosts and routers into an IPv4
world (with the long-term goal, of course, of having all IPv4 nodes eventually transition to IPv6).
Probably the most straightforward way to introduce IPv6-capable nodes is a
dual-stack approach, where IPv6 nodes also have a complete IPv4 implementation.
Such a node, referred to as an IPv6/IPv4 node in RFC 4213, has the ability to send
and receive both IPv4 and IPv6 datagrams. When interoperating with an IPv4 node,
an IPv6/IPv4 node can use IPv4 datagrams; when interoperating with an IPv6 node,
it can speak IPv6. IPv6/IPv4 nodes must have both IPv6 and IPv4 addresses. They
359
360
CHAPTER 4
•
THE NETWORK LAYER
must furthermore be able to determine whether another node is IPv6-capable or
IPv4-only. This problem can be solved using the DNS (see Chapter 2), which can
return an IPv6 address if the node name being resolved is IPv6-capable, or otherwise return an IPv4 address. Of course, if the node issuing the DNS request is only
IPv4-capable, the DNS returns only an IPv4 address.
In the dual-stack approach, if either the sender or the receiver is only IPv4capable, an IPv4 datagram must be used. As a result, it is possible that two IPv6capable nodes can end up, in essence, sending IPv4 datagrams to each other. This is
illustrated in Figure 4.25. Suppose Node A is IPv6-capable and wants to send an IP
datagram to Node F, which is also IPv6-capable. Nodes A and B can exchange an
IPv6 datagram. However, Node B must create an IPv4 datagram to send to C. Certainly, the data field of the IPv6 datagram can be copied into the data field of the
IPv4 datagram and appropriate address mapping can be done. However, in performing the conversion from IPv6 to IPv4, there will be IPv6-specific fields in the IPv6
datagram (for example, the flow identifier field) that have no counterpart in IPv4.
The information in these fields will be lost. Thus, even though E and F can exchange
IPv6 datagrams, the arriving IPv4 datagrams at E from D do not contain all of the
fields that were in the original IPv6 datagram sent from A.
An alternative to the dual-stack approach, also discussed in RFC 4213, is
known as tunneling. Tunneling can solve the problem noted above, allowing, for
example, E to receive the IPv6 datagram originated by A. The basic idea behind
tunneling is the following. Suppose two IPv6 nodes (for example, B and E in Figure 4.25) want to interoperate using IPv6 datagrams but are connected to each
other by intervening IPv4 routers. We refer to the intervening set of IPv4 routers
between two IPv6 routers as a tunnel, as illustrated in Figure 4.26. With tunneling, the IPv6 node on the sending side of the tunnel (for example, B) takes the
entire IPv6 datagram and puts it in the data (payload) field of an IPv4 datagram.
IPv6
IPv6
IPv4
IPv4
IPv6
IPv6
A
B
C
D
E
F
Flow: X
Source: A
Dest: F
Source: A
Dest: F
Source: A
Dest: F
data
data
data
data
A to B: IPv6
Figure 4.25
Flow: ??
Source: A
Dest: F
B to C: IPv4
A dual-stack approach
D to E: IPv4
E to F: IPv6
4.4
•
THE INTERNET PROTOCOL (IP)
Logical view
IPv6
IPv6
A
B
Tunnel
IPv6
IPv6
E
F
Physical view
IPv6
IPv6
IPv4
IPv4
IPv6
IPv6
A
B
C
D
E
F
Flow: X
Source: A
Dest: F
Source: B
Dest: E
data
Source: B
Dest: E
Flow: X
Source: A
Dest: F
Flow: X
Source: A
Dest: F
data
data
A to B: IPv6
data
E to F: IPv6
B to C: IPv4
(encapsulating IPv6)
Figure 4.26
Flow: X
Source: A
Dest: F
D to E: IPv4
(encapsulating IPv6)
Tunneling
This IPv4 datagram is then addressed to the IPv6 node on the receiving side of
the tunnel (for example, E) and sent to the first node in the tunnel (for example,
C). The intervening IPv4 routers in the tunnel route this IPv4 datagram among
themselves, just as they would any other datagram, blissfully unaware that the
IPv4 datagram itself contains a complete IPv6 datagram. The IPv6 node on the
receiving side of the tunnel eventually receives the IPv4 datagram (it is the destination of the IPv4 datagram!), determines that the IPv4 datagram contains an
IPv6 datagram, extracts the IPv6 datagram, and then routes the IPv6 datagram
exactly as it would if it had received the IPv6 datagram from a directly connected
IPv6 neighbor.
We end this section by noting that while the adoption of IPv6 was initially
slow to take off [Lawton 2001], momentum has been building recently. See [Huston 2008b] for discussion of IPv6 deployment as of 2008; see [NIST IPv6 2012]
for a snapshort of US IPv6 deployment. The proliferation of devices such as IPenabled phones and other portable devices provides an additional push for more
361
362
CHAPTER 4
•
THE NETWORK LAYER
widespread deployment of IPv6. Europe’s Third Generation Partnership Program
[3GPP 2012] has specified IPv6 as the standard addressing scheme for mobile
multimedia.
One important lesson that we can learn from the IPv6 experience is that it is enormously difficult to change network-layer protocols. Since the early 1990s, numerous
new network-layer protocols have been trumpeted as the next major revolution for the
Internet, but most of these protocols have had limited penetration to date. These protocols include IPv6, multicast protocols (Section 4.7), and resource reservation protocols (Chapter 7). Indeed, introducing new protocols into the network layer is like
replacing the foundation of a house—it is difficult to do without tearing the whole
house down or at least temporarily relocating the house’s residents. On the other hand,
the Internet has witnessed rapid deployment of new protocols at the application layer.
The classic examples, of course, are the Web, instant messaging, and P2P file sharing.
Other examples include audio and video streaming and distributed games. Introducing
new application-layer protocols is like adding a new layer of paint to a house—it is
relatively easy to do, and if you choose an attractive color, others in the neighborhood
will copy you. In summary, in the future we can expect to see changes in the Internet’s
network layer, but these changes will likely occur on a time scale that is much slower
than the changes that will occur at the application layer.
4.4.5 A Brief Foray into IP Security
Section 4.4.3 covered IPv4 in some detail, including the services it provides and
how those services are implemented. While reading through that section, you may
have noticed that there was no mention of any security services. Indeed, IPv4 was
designed in an era (the 1970s) when the Internet was primarily used among mutually-trusted networking researchers. Creating a computer network that integrated a
multitude of link-layer technologies was already challenging enough, without having to worry about security.
But with security being a major concern today, Internet researchers have moved
on to design new network-layer protocols that provide a variety of security services.
One of these protocols is IPsec, one of the more popular secure network-layer protocols and also widely deployed in Virtual Private Networks (VPNs). Although IPsec and
its cryptographic underpinnings are covered in some detail in Chapter 8, we provide a
brief, high-level introduction into IPsec services in this section.
IPsec has been designed to be backward compatible with IPv4 and IPv6. In particular, in order to reap the benefits of IPsec, we don’t need to replace the protocol
stacks in all the routers and hosts in the Internet. For example, using the transport
mode (one of two IPsec “modes”), if two hosts want to securely communicate, IPsec
needs to be available only in those two hosts. All other routers and hosts can continue to run vanilla IPv4.
For concreteness, we’ll focus on IPsec’s transport mode here. In this mode, two
hosts first establish an IPsec session between themselves. (Thus IPsec is connectionoriented!) With the session in place, all TCP and UDP segments sent between the
4.5
•
ROUTING ALGORITHMS
two hosts enjoy the security services provided by IPsec. On the sending side, the
transport layer passes a segment to IPsec. IPsec then encrypts the segment, appends
additional security fields to the segment, and encapsulates the resulting payload in
an ordinary IP datagram. (It’s actually a little more complicated than this, as we’ll
see in Chapter 8.) The sending host then sends the datagram into the Internet, which
transports it to the destination host. There, IPsec decrypts the segment and passes
the unencrypted segment to the transport layer.
The services provided by an IPsec session include:
• Cryptographic agreement. Mechanisms that allow the two communicating hosts
to agree on cryptographic algorithms and keys.
• Encryption of IP datagram payloads. When the sending host receives a segment
from the transport layer, IPsec encrypts the payload. The payload can only be
decrypted by IPsec in the receiving host.
• Data integrity. IPsec allows the receiving host to verify that the datagram’s
header fields and encrypted payload were not modified while the datagram was
en route from source to destination.
• Origin authentication. When a host receives an IPsec datagram from a trusted
source (with a trusted key—see Chapter 8), the host is assured that the source IP
address in the datagram is the actual source of the datagram.
When two hosts have an IPsec session established between them, all TCP and
UDP segments sent between them will be encrypted and authenticated. IPsec therefore provides blanket coverage, securing all communication between the two hosts
for all network applications.
A company can use IPsec to communicate securely in the nonsecure public Internet. For illustrative purposes, we’ll just look at a simple example here. Consider a
company that has a large number of traveling salespeople, each possessing a company
laptop computer. Suppose the salespeople need to frequently consult sensitive company information (for example, pricing and product information) that is stored on a
server in the company’s headquarters. Further suppose that the salespeople also need
to send sensitive documents to each other. How can this be done with IPsec? As you
might guess, we install IPsec in the server and in all of the salespeople’s laptops. With
IPsec installed in these hosts, whenever a salesperson needs to communicate with the
server or with another salesperson, the communication session will be secure.
4.5 Routing Algorithms
So far in this chapter, we’ve mostly explored the network layer’s forwarding function. We learned that when a packet arrives to a router, the router indexes a forwarding table and determines the link interface to which the packet is to be directed. We
also learned that routing algorithms, operating in network routers, exchange and
363
364
CHAPTER 4
•
THE NETWORK LAYER
compute the information that is used to configure these forwarding tables. The interplay between routing algorithms and forwarding tables was shown in Figure 4.2.
Having explored forwarding in some depth we now turn our attention to the other
major topic of this chapter, namely, the network layer’s critical routing function.
Whether the network layer provides a datagram service (in which case different packets between a given source-destination pair may take different routes) or a VC service (in which case all packets between a given source and destination will take the
same path), the network layer must nonetheless determine the path that packets take
from senders to receivers. We’ll see that the job of routing is to determine good paths
(equivalently, routes), from senders to receivers, through the network of routers.
Typically a host is attached directly to one router, the default router for the
host (also called the first-hop router for the host). Whenever a host sends a packet,
the packet is transferred to its default router. We refer to the default router of the
source host as the source router and the default router of the destination host as the
destination router. The problem of routing a packet from source host to destination
host clearly boils down to the problem of routing the packet from source router to
destination router, which is the focus of this section.
The purpose of a routing algorithm is then simple: given a set of routers, with
links connecting the routers, a routing algorithm finds a “good” path from source
router to destination router. Typically, a good path is one that has the least cost.
We’ll see, however, that in practice, real-world concerns such as policy issues (for
example, a rule such as “router x, belonging to organization Y, should not forward
any packets originating from the network owned by organization Z”) also come into
play to complicate the conceptually simple and elegant algorithms whose theory
underlies the practice of routing in today’s networks.
A graph is used to formulate routing problems. Recall that a graph G = (N,E)
is a set N of nodes and a collection E of edges, where each edge is a pair of nodes
from N. In the context of network-layer routing, the nodes in the graph represent
routers—the points at which packet-forwarding decisions are made—and the edges
connecting these nodes represent the physical links between these routers. Such a
graph abstraction of a computer network is shown in Figure 4.27. To view some
graphs representing real network maps, see [Dodge 2012, Cheswick 2000]; for a
discussion of how well different graph-based models model the Internet, see
[Zegura 1997, Faloutsos 1999, Li 2004].
As shown in Figure 4.27, an edge also has a value representing its cost. Typically, an edge’s cost may reflect the physical length of the corresponding link (for
example, a transoceanic link might have a higher cost than a short-haul terrestrial
link), the link speed, or the monetary cost associated with a link. For our purposes,
we’ll simply take the edge costs as a given and won’t worry about how they are
determined. For any edge (x,y) in E, we denote c(x,y) as the cost of the edge between
nodes x and y. If the pair (x,y) does not belong to E, we set c(x,y) = ∞. Also, throughout we consider only undirected graphs (i.e., graphs whose edges do not have a
direction), so that edge (x,y) is the same as edge (y,x) and that c(x,y) = c(y,x). Also, a
node y is said to be a neighbor of node x if (x,y) belongs to E.
4.5
•
ROUTING ALGORITHMS
5
2
v
2
u
3
w
5
3
1
1
x
Figure 4.27
z
2
1
y
Abstract graph model of a computer network
Given that costs are assigned to the various edges in the graph abstraction, a natural goal of a routing algorithm is to identify the least costly paths between sources and
destinations. To make this problem more precise, recall that a path in a graph G =
(N,E) is a sequence of nodes (x1, x2,..., xp) such that each of the pairs (x1,x2),
(x2,x3),...,(xp-1,xp) are edges in E. The cost of a path (x1,x2,..., xp) is simply the sum of
all the edge costs along the path, that is, c(x1,x2) + c(x2,x3) + ...+ c(xp-1,xp). Given any
two nodes x and y, there are typically many paths between the two nodes, with each
path having a cost. One or more of these paths is a least-cost path. The least-cost
problem is therefore clear: Find a path between the source and destination that has
least cost. In Figure 4.27, for example, the least-cost path between source node u and
destination node w is (u, x, y, w) with a path cost of 3. Note that if all edges in the
graph have the same cost, the least-cost path is also the shortest path (that is, the
path with the smallest number of links between the source and the destination).
As a simple exercise, try finding the least-cost path from node u to z in Figure
4.27 and reflect for a moment on how you calculated that path. If you are like most
people, you found the path from u to z by examining Figure 4.27, tracing a few routes
from u to z, and somehow convincing yourself that the path you had chosen had the
least cost among all possible paths. (Did you check all of the 17 possible paths
between u and z? Probably not!) Such a calculation is an example of a centralized
routing algorithm—the routing algorithm was run in one location, your brain, with
complete information about the network. Broadly, one way in which we can classify
routing algorithms is according to whether they are global or decentralized.
• A global routing algorithm computes the least-cost path between a source and
destination using complete, global knowledge about the network. That is, the
algorithm takes the connectivity between all nodes and all link costs as inputs.
This then requires that the algorithm somehow obtain this information before
actually performing the calculation. The calculation itself can be run at one site
365
366
CHAPTER 4
•
THE NETWORK LAYER
(a centralized global routing algorithm) or replicated at multiple sites. The key
distinguishing feature here, however, is that a global algorithm has complete
information about connectivity and link costs. In practice, algorithms with global
state information are often referred to as link-state (LS) algorithms, since the
algorithm must be aware of the cost of each link in the network. We’ll study LS
algorithms in Section 4.5.1.
• In a decentralized routing algorithm, the calculation of the least-cost path is
carried out in an iterative, distributed manner. No node has complete information
about the costs of all network links. Instead, each node begins with only the
knowledge of the costs of its own directly attached links. Then, through an iterative process of calculation and exchange of information with its neighboring
nodes (that is, nodes that are at the other end of links to which it itself is
attached), a node gradually calculates the least-cost path to a destination or set of
destinations. The decentralized routing algorithm we’ll study below in Section
4.5.2 is called a distance-vector (DV) algorithm, because each node maintains a
vector of estimates of the costs (distances) to all other nodes in the network.
A second broad way to classify routing algorithms is according to whether they
are static or dynamic. In static routing algorithms, routes change very slowly over
time, often as a result of human intervention (for example, a human manually editing a router’s forwarding table). Dynamic routing algorithms change the routing
paths as the network traffic loads or topology change. A dynamic algorithm can be
run either periodically or in direct response to topology or link cost changes. While
dynamic algorithms are more responsive to network changes, they are also more
susceptible to problems such as routing loops and oscillation in routes.
A third way to classify routing algorithms is according to whether they are loadsensitive or load-insensitive. In a load-sensitive algorithm, link costs vary dynamically to reflect the current level of congestion in the underlying link. If a high cost is
associated with a link that is currently congested, a routing algorithm will tend to
choose routes around such a congested link. While early ARPAnet routing algorithms were load-sensitive [McQuillan 1980], a number of difficulties were encountered [Huitema 1998]. Today’s Internet routing algorithms (such as RIP, OSPF, and
BGP) are load-insensitive, as a link’s cost does not explicitly reflect its current (or
recent past) level of congestion.
4.5.1 The Link-State (LS) Routing Algorithm
Recall that in a link-state algorithm, the network topology and all link costs are
known, that is, available as input to the LS algorithm. In practice this is accomplished by having each node broadcast link-state packets to all other nodes in the
network, with each link-state packet containing the identities and costs of its
attached links. In practice (for example, with the Internet’s OSPF routing protocol,
discussed in Section 4.6.1) this is often accomplished by a link-state broadcast
4.5
•
ROUTING ALGORITHMS
algorithm [Perlman 1999]. We’ll cover broadcast algorithms in Section 4.7. The
result of the nodes’ broadcast is that all nodes have an identical and complete view
of the network. Each node can then run the LS algorithm and compute the same set
of least-cost paths as every other node.
The link-state routing algorithm we present below is known as Dijkstra’s algorithm, named after its inventor. A closely related algorithm is Prim’s algorithm; see
[Cormen 2001] for a general discussion of graph algorithms. Dijkstra’s algorithm
computes the least-cost path from one node (the source, which we will refer to as u)
to all other nodes in the network. Dijkstra’s algorithm is iterative and has the property that after the kth iteration of the algorithm, the least-cost paths are known to k
destination nodes, and among the least-cost paths to all destination nodes, these k
paths will have the k smallest costs. Let us define the following notation:
• D(v): cost of the least-cost path from the source node to destination v as of this
iteration of the algorithm.
• p(v): previous node (neighbor of v) along the current least-cost path from the
source to v.
• N⬘ : subset of nodes; v is in N⬘ if the least-cost path from the source to v is definitively known.
The global routing algorithm consists of an initialization step followed by a
loop. The number of times the loop is executed is equal to the number of nodes in
the network. Upon termination, the algorithm will have calculated the shortest paths
from the source node u to every other node in the network.
Link-State (LS) Algorithm for Source Node u
1 Initialization:
2
N’ = {u}
3
for all nodes v
4
if v is a neighbor of u
5
then D(v) = c(u,v)
6
else D(v) = ∞
7
8 Loop
9
find w not in N’ such that D(w) is a minimum
10
add w to N’
11
update D(v) for each neighbor v of w and not in N’:
12
D(v) = min( D(v), D(w) + c(w,v) )
13
/* new cost to v is either old cost to v or known
14
least path cost to w plus cost from w to v */
15 until N’= N
367
368
CHAPTER 4
VideoNote
Dijkstra’s algorithm:
discussion and example
•
THE NETWORK LAYER
As an example, let’s consider the network in Figure 4.27 and compute the
least-cost paths from u to all possible destinations. A tabular summary of the
algorithm’s computation is shown in Table 4.3, where each line in the table gives
the values of the algorithm’s variables at the end of the iteration. Let’s consider
the few first steps in detail.
• In the initialization step, the currently known least-cost paths from u to its
directly attached neighbors, v, x, and w, are initialized to 2, 1, and 5, respectively.
Note in particular that the cost to w is set to 5 (even though we will soon see that
a lesser-cost path does indeed exist) since this is the cost of the direct (one hop)
link from u to w. The costs to y and z are set to infinity because they are not
directly connected to u.
• In the first iteration, we look among those nodes not yet added to the set N⬘ and
find that node with the least cost as of the end of the previous iteration. That node
is x, with a cost of 1, and thus x is added to the set N⬘. Line 12 of the LS algorithm is then performed to update D(v) for all nodes v, yielding the results shown
in the second line (Step 1) in Table 4.3. The cost of the path to v is unchanged.
The cost of the path to w (which was 5 at the end of the initialization) through
node x is found to have a cost of 4. Hence this lower-cost path is selected and w’s
predecessor along the shortest path from u is set to x. Similarly, the cost to y
(through x) is computed to be 2, and the table is updated accordingly.
• In the second iteration, nodes v and y are found to have the least-cost paths (2),
and we break the tie arbitrarily and add y to the set N⬘ so that N⬘ now contains u,
x, and y. The cost to the remaining nodes not yet in N⬘, that is, nodes v, w, and z,
are updated via line 12 of the LS algorithm, yielding the results shown in the
third row in the Table 4.3.
• And so on. . . .
When the LS algorithm terminates, we have, for each node, its predecessor
along the least-cost path from the source node. For each predecessor, we also
step
0
1
2
3
4
5
N’
u
ux
uxy
uxyv
uxyvw
uxyvwz
Table 4.3
D(v),p(v)
D(w),p(w)
D(x),p(x)
D(y),p(y)
D(z),p(z)
2,u
2,u
2,u
5,u
4,x
3,y
3,y
1,u
∞
2,x
∞
∞
4,y
4,y
4,y
Running the link-state algorithm on the network in Figure 4.27
4.5
•
ROUTING ALGORITHMS
have its predecessor, and so in this manner we can construct the entire path from
the source to all destinations. The forwarding table in a node, say node u, can
then be constructed from this information by storing, for each destination, the
next-hop node on the least-cost path from u to the destination. Figure 4.28
shows the resulting least-cost paths and forwarding table in u for the network in
Figure 4.27.
What is the computational complexity of this algorithm? That is, given n
nodes (not counting the source), how much computation must be done in the
worst case to find the least-cost paths from the source to all destinations? In the
first iteration, we need to search through all n nodes to determine the node, w, not
in N⬘ that has the minimum cost. In the second iteration, we need to check n – 1
nodes to determine the minimum cost; in the third iteration n – 2 nodes, and so
on. Overall, the total number of nodes we need to search through over all the iterations is n(n + 1)/2, and thus we say that the preceding implementation of the LS
algorithm has worst-case complexity of order n squared: O(n2). (A more sophisticated implementation of this algorithm, using a data structure known as a heap,
can find the minimum in line 9 in logarithmic rather than linear time, thus reducing the complexity.)
Before completing our discussion of the LS algorithm, let us consider a pathology that can arise. Figure 4.29 shows a simple network topology where link costs
are equal to the load carried on the link, for example, reflecting the delay that would
be experienced. In this example, link costs are not symmetric; that is, c(u,v) equals
c(v,u) only if the load carried on both directions on the link (u,v) is the same. In this
example, node z originates a unit of traffic destined for w, node x also originates a
unit of traffic destined for w, and node y injects an amount of traffic equal to e, also
destined for w. The initial routing is shown in Figure 4.29(a) with the link costs corresponding to the amount of traffic carried.
When the LS algorithm is next run, node y determines (based on the link costs
shown in Figure 4.29(a)) that the clockwise path to w has a cost of 1, while the
counterclockwise path to w (which it had been using) has a cost of 1 + e. Hence y’s
V
W
Z
U
X
Figure 4.28
Y
Destination
Link
v
w
x
y
z
(u, v)
(u, x)
(u, x)
(u, x)
(u, x)
Least cost path and forwarding table for nodule u
369
370
CHAPTER 4
•
THE NETWORK LAYER
w
w
1+e
0
1
1
2+e
z
x
0
1
1
z
0
1+e
e
0
0
y
e
e
0
0
2+e
z
x
0
1
1
z
1+e
0
1
1+e
0
1
0
y
y
e
e
c. x, y, z detect better path
to w, counterclockwise
1
w
2+ e
Figure 4.29
x
b. x, y detect better path
to w, clockwise
w
1
1
0
y
a. Initial routing
x
1
d. x, y, z, detect better path
to w, clockwise
Oscillations with congestion-sensitive routing
least-cost path to w is now clockwise. Similarly, x determines that its new least-cost
path to w is also clockwise, resulting in costs shown in Figure 4.29(b). When the
LS algorithm is run next, nodes x, y, and z all detect a zero-cost path to w in the
counterclockwise direction, and all route their traffic to the counterclockwise
routes. The next time the LS algorithm is run, x, y, and z all then route their traffic
to the clockwise routes.
What can be done to prevent such oscillations (which can occur in any algorithm, not just an LS algorithm, that uses a congestion or delay-based link metric)? One solution would be to mandate that link costs not depend on the amount
of traffic carried—an unacceptable solution since one goal of routing is to avoid
4.5
•
ROUTING ALGORITHMS
highly congested (for example, high-delay) links. Another solution is to ensure
that not all routers run the LS algorithm at the same time. This seems a more
reasonable solution, since we would hope that even if routers ran the LS algorithm
with the same periodicity, the execution instance of the algorithm would not be
the same at each node. Interestingly, researchers have found that routers in the
Internet can self-synchronize among themselves [Floyd Synchronization 1994].
That is, even though they initially execute the algorithm with the same period
but at different instants of time, the algorithm execution instance can eventually
become, and remain, synchronized at the routers. One way to avoid such selfsynchronization is for each router to randomize the time it sends out a link
advertisement.
Having studied the LS algorithm, let’s consider the other major routing algorithm that is used in practice today—the distance-vector routing algorithm.
4.5.2 The Distance-Vector (DV) Routing Algorithm
Whereas the LS algorithm is an algorithm using global information, the distancevector (DV) algorithm is iterative, asynchronous, and distributed. It is distributed
in that each node receives some information from one or more of its directly
attached neighbors, performs a calculation, and then distributes the results of its
calculation back to its neighbors. It is iterative in that this process continues
on until no more information is exchanged between neighbors. (Interestingly, the
algorithm is also self-terminating—there is no signal that the computation should
stop; it just stops.) The algorithm is asynchronous in that it does not require all of
the nodes to operate in lockstep with each other. We’ll see that an asynchronous,
iterative, self-terminating, distributed algorithm is much more interesting and fun
than a centralized algorithm!
Before we present the DV algorithm, it will prove beneficial to discuss an
important relationship that exists among the costs of the least-cost paths. Let dx(y)
be the cost of the least-cost path from node x to node y. Then the least costs are
related by the celebrated Bellman-Ford equation, namely,
dx(y) = minv{c(x,v) + dv(y)},
(4.1)
where the minv in the equation is taken over all of x’s neighbors. The Bellman-Ford
equation is rather intuitive. Indeed, after traveling from x to v, if we then take the
least-cost path from v to y, the path cost will be c(x,v) + dv(y). Since we must begin
by traveling to some neighbor v, the least cost from x to y is the minimum of c(x,v)
+ dv(y) taken over all neighbors v.
But for those who might be skeptical about the validity of the equation, let’s
check it for source node u and destination node z in Figure 4.27. The source node u
371
372
CHAPTER 4
•
THE NETWORK LAYER
has three neighbors: nodes v, x, and w. By walking along various paths in the graph,
it is easy to see that dv(z) = 5, dx(z) = 3, and dw(z) = 3. Plugging these values into
Equation 4.1, along with the costs c(u,v) = 2, c(u,x) = 1, and c(u,w) = 5, gives du(z) =
min{2 + 5, 5 + 3, 1 + 3} = 4, which is obviously true and which is exactly what the
Dijskstra algorithm gave us for the same network. This quick verification should
help relieve any skepticism you may have.
The Bellman-Ford equation is not just an intellectual curiosity. It actually has
significant practical importance. In particular, the solution to the Bellman-Ford
equation provides the entries in node x’s forwarding table. To see this, let v* be any
neighboring node that achieves the minimum in Equation 4.1. Then, if node x wants
to send a packet to node y along a least-cost path, it should first forward the packet
to node v*. Thus, node x’s forwarding table would specify node v* as the next-hop
router for the ultimate destination y. Another important practical contribution of the
Bellman-Ford equation is that it suggests the form of the neighbor-to-neighbor communication that will take place in the DV algorithm.
The basic idea is as follows. Each node x begins with Dx(y), an estimate of the
cost of the least-cost path from itself to node y, for all nodes in N. Let Dx = [Dx(y): y
in N] be node x’s distance vector, which is the vector of cost estimates from x to all
other nodes, y, in N. With the DV algorithm, each node x maintains the following
routing information:
• For each neighbor v, the cost c(x,v) from x to directly attached neighbor, v
• Node x’s distance vector, that is, Dx = [Dx(y): y in N], containing x’s estimate of
its cost to all destinations, y, in N
• The distance vectors of each of its neighbors, that is, Dv = [Dv(y): y in N] for each
neighbor v of x
In the distributed, asynchronous algorithm, from time to time, each node sends
a copy of its distance vector to each of its neighbors. When a node x receives a
new distance vector from any of its neighbors v, it saves v’s distance vector, and
then uses the Bellman-Ford equation to update its own distance vector as follows:
Dx(y) ⫽ minv{c(x,v) + Dv(y)}
for each node y in N
If node x’s distance vector has changed as a result of this update step, node x will
then send its updated distance vector to each of its neighbors, which can in turn
update their own distance vectors. Miraculously enough, as long as all the nodes
continue to exchange their distance vectors in an asynchronous fashion, each cost
estimate Dx(y) converges to dx(y), the actual cost of the least-cost path from node x
to node y [Bertsekas 1991]!
4.5
•
ROUTING ALGORITHMS
373
Distance-Vector (DV) Algorithm
At each node, x:
1 Initialization:
2
for all destinations y in N:
3
Dx(y) = c(x,y)
/* if y is not a neighbor then c(x,y) = ∞ */
4
for each neighbor w
5
Dw(y) = ? for all destinations y in N
6
for each neighbor w
7
send distance vector Dx = [Dx(y): y in N] to w
8
9 loop
10
wait (until I see a link cost change to some neighbor w or
11
until I receive a distance vector from some neighbor w)
12
13
for each y in N:
14
Dx(y) = minv{c(x,v) + Dv(y)}
15
16
if Dx(y) changed for any destination y
17
send distance vector Dx = [Dx(y): y in N] to all neighbors
18
19 forever
In the DV algorithm, a node x updates its distance-vector estimate when it
either sees a cost change in one of its directly attached links or receives a distancevector update from some neighbor. But to update its own forwarding table for a
given destination y, what node x really needs to know is not the shortest-path
distance to y but instead the neighboring node v*(y) that is the next-hop router along
the shortest path to y. As you might expect, the next-hop router v*(y) is the neighbor
v that achieves the minimum in Line 14 of the DV algorithm. (If there are multiple
neighbors v that achieve the minimum, then v*(y) can be any of the minimizing
neighbors.) Thus, in Lines 13–14, for each destination y, node x also determines
v*(y) and updates its forwarding table for destination y.
Recall that the LS algorithm is a global algorithm in the sense that it requires
each node to first obtain a complete map of the network before running the Dijkstra
algorithm. The DV algorithm is decentralized and does not use such global information. Indeed, the only information a node will have is the costs of the links to its
directly attached neighbors and information it receives from these neighbors. Each
node waits for an update from any neighbor (Lines 10–11), calculates its new distance vector when receiving an update (Line 14), and distributes its new distance
•
THE NETWORK LAYER
vector to its neighbors (Lines 16–17). DV-like algorithms are used in many routing
protocols in practice, including the Internet’s RIP and BGP, ISO IDRP, Novell IPX,
and the original ARPAnet.
Figure 4.30 illustrates the operation of the DV algorithm for the simple threenode network shown at the top of the figure. The operation of the algorithm is illustrated in a synchronous manner, where all nodes simultaneously receive distance
vectors from their neighbors, compute their new distance vectors, and inform their
neighbors if their distance vectors have changed. After studying this example, you
y
2
1
x
7
z
Node x table
0 2 7
∞ ∞ ∞
∞ ∞ ∞
x
y
z
0 2 3
2 0 1
7 1 0
cost to
x y z
from
x
y
z
cost to
x y z
from
from
cost to
x y z
x
y
z
0 2 3
2 0 1
3 1 0
Node y table
∞ ∞ ∞
2 0 1
∞ ∞ ∞
x
y
z
0 2 7
2 0 1
7 1 0
cost to
x y z
from
x
y
z
cost to
x y z
from
cost to
x y z
from
x
y
z
0 2 3
2 0 1
3 1 0
Node z table
x
y
z
∞ ∞ ∞
∞ ∞ ∞
7 1 0
cost to
x y z
x
y
z
0 2 7
2 0 1
3 1 0
cost to
x y z
from
cost to
x y z
from
CHAPTER 4
from
374
x
y
z
0 2 3
2 0 1
3 1 0
Time
Figure 4.30
Distance-vector (DV) algorithm
4.5
•
ROUTING ALGORITHMS
should convince yourself that the algorithm operates correctly in an asynchronous
manner as well, with node computations and update generation/reception occurring
at any time.
The leftmost column of the figure displays three initial routing tables for each
of the three nodes. For example, the table in the upper-left corner is node x’s initial
routing table. Within a specific routing table, each row is a distance vector—specifically, each node’s routing table includes its own distance vector and that of each of
its neighbors. Thus, the first row in node x’s initial routing table is Dx = [Dx(x),
Dx(y), Dx(z)] = [0, 2, 7]. The second and third rows in this table are the most recently
received distance vectors from nodes y and z, respectively. Because at initialization
node x has not received anything from node y or z, the entries in the second and third
rows are initialized to infinity.
After initialization, each node sends its distance vector to each of its two neighbors. This is illustrated in Figure 4.30 by the arrows from the first column of tables
to the second column of tables. For example, node x sends its distance vector Dx =
[0, 2, 7] to both nodes y and z. After receiving the updates, each node recomputes its
own distance vector. For example, node x computes
Dx(x) = 0
Dx(y) = min{c(x,y) + Dy(y), c(x,z) + Dz(y)} = min{2 + 0, 7 + 1} = 2
Dx(z) = min{c(x,y) + Dy(z), c(x,z) + Dz(z)} = min{2 + 1, 7 + 0} = 3
The second column therefore displays, for each node, the node’s new distance vector along with distance vectors just received from its neighbors. Note, for example,
that node x’s estimate for the least cost to node z, Dx(z), has changed from 7 to 3.
Also note that for node x, neighboring node y achieves the minimum in line 14 of
the DV algorithm; thus at this stage of the algorithm, we have at node x that v*(y) =
y and v*(z) = y.
After the nodes recompute their distance vectors, they again send their updated
distance vectors to their neighbors (if there has been a change). This is illustrated in
Figure 4.30 by the arrows from the second column of tables to the third column of
tables. Note that only nodes x and z send updates: node y’s distance vector didn’t
change so node y doesn’t send an update. After receiving the updates, the nodes then
recompute their distance vectors and update their routing tables, which are shown in
the third column.
The process of receiving updated distance vectors from neighbors, recomputing
routing table entries, and informing neighbors of changed costs of the least-cost path
to a destination continues until no update messages are sent. At this point, since no
update messages are sent, no further routing table calculations will occur and the
algorithm will enter a quiescent state; that is, all nodes will be performing the wait
in Lines 10–11 of the DV algorithm. The algorithm remains in the quiescent state
until a link cost changes, as discussed next.
375
376
CHAPTER 4
•
THE NETWORK LAYER
Distance-Vector Algorithm: Link-Cost Changes and Link Failure
When a node running the DV algorithm detects a change in the link cost from itself to
a neighbor (Lines 10–11), it updates its distance vector (Lines 13–14) and, if there’s a
change in the cost of the least-cost path, informs its neighbors (Lines 16–17) of its new
distance vector. Figure 4.31(a) illustrates a scenario where the link cost from y to x
changes from 4 to 1. We focus here only on y’ and z’s distance table entries to destination x. The DV algorithm causes the following sequence of events to occur:
• At time t0, y detects the link-cost change (the cost has changed from 4 to 1),
updates its distance vector, and informs its neighbors of this change since its distance vector has changed.
• At time t1, z receives the update from y and updates its table. It computes a new
least cost to x (it has decreased from a cost of 5 to a cost of 2) and sends its new
distance vector to its neighbors.
• At time t2, y receives z’s update and updates its distance table. y’s least costs do
not change and hence y does not send any message to z. The algorithm comes to
a quiescent state.
Thus, only two iterations are required for the DV algorithm to reach a quiescent
state. The good news about the decreased cost between x and y has propagated
quickly through the network.
Let’s now consider what can happen when a link cost increases. Suppose that
the link cost between x and y increases from 4 to 60, as shown in Figure 4.31(b).
1. Before the link cost changes, Dy(x) = 4, Dy(z) = 1, Dz(y) = 1, and Dz(x) = 5. At
time t0, y detects the link-cost change (the cost has changed from 4 to 60). y
computes its new minimum-cost path to x to have a cost of
Dy(x) = min{c(y,x) + Dx(x), c(y,z) + Dz(x)} = min{60 + 0, 1 + 5} = 6
1
60
y
4
x
50
x
z
a.
Figure 4.31
b.
y
4
1
Changes in link cost
1
50
z
4.5
•
ROUTING ALGORITHMS
Of course, with our global view of the network, we can see that this new cost
via z is wrong. But the only information node y has is that its direct cost to x is
60 and that z has last told y that z could get to x with a cost of 5. So in order to
get to x, y would now route through z, fully expecting that z will be able to get
to x with a cost of 5. As of t1 we have a routing loop—in order to get to x, y
routes through z, and z routes through y. A routing loop is like a black hole—a
packet destined for x arriving at y or z as of t1 will bounce back and forth
between these two nodes forever (or until the forwarding tables are changed).
2. Since node y has computed a new minimum cost to x, it informs z of its new
distance vector at time t1.
3. Sometime after t1, z receives y’s new distance vector, which indicates that y’s
minimum cost to x is 6. z knows it can get to y with a cost of 1 and hence
computes a new least cost to x of Dz(x) = min{50 + 0,1 + 6} = 7. Since z’s
least cost to x has increased, it then informs y of its new distance vector at t2.
4. In a similar manner, after receiving z’s new distance vector, y determines
Dy(x) = 8 and sends z its distance vector. z then determines Dz(x) = 9 and
sends y its distance vector, and so on.
How long will the process continue? You should convince yourself that the loop
will persist for 44 iterations (message exchanges between y and z)—until z eventually computes the cost of its path via y to be greater than 50. At this point, z will
(finally!) determine that its least-cost path to x is via its direct connection to x. y
will then route to x via z. The result of the bad news about the increase in link
cost has indeed traveled slowly! What would have happened if the link cost c(y,
x) had changed from 4 to 10,000 and the cost c(z, x) had been 9,999? Because of
such scenarios, the problem we have seen is sometimes referred to as the countto-infinity problem.
Distance-Vector Algorithm: Adding Poisoned Reverse
The specific looping scenario just described can be avoided using a technique
known as poisoned reverse. The idea is simple—if z routes through y to get to
destination x, then z will advertise to y that its distance to x is infinity, that is, z will
advertise to y that Dz(x) = ∞ (even though z knows Dz(x) = 5 in truth). z will continue telling this little white lie to y as long as it routes to x via y. Since y believes
that z has no path to x, y will never attempt to route to x via z, as long as z continues
to route to x via y (and lies about doing so).
Let’s now see how poisoned reverse solves the particular looping problem we
encountered before in Figure 4.31(b). As a result of the poisoned reverse, y’s distance table indicates Dz(x) = ∞. When the cost of the (x, y) link changes from 4 to 60
at time t0, y updates its table and continues to route directly to x, albeit at a higher
cost of 60, and informs z of its new cost to x, that is, Dy(x) = 60. After receiving the
377
378
CHAPTER 4
•
THE NETWORK LAYER
update at t1, z immediately shifts its route to x to be via the direct (z, x) link at a cost
of 50. Since this is a new least-cost path to x, and since the path no longer passes
through y, z now informs y that Dz(x) = 50 at t2. After receiving the update from z, y
updates its distance table with Dy(x) = 51. Also, since z is now on y’s least-cost path
to x, y poisons the reverse path from z to x by informing z at time t3 that Dy(x) = ∞
(even though y knows that Dy(x) = 51 in truth).
Does poisoned reverse solve the general count-to-infinity problem? It does not.
You should convince yourself that loops involving three or more nodes (rather than
simply two immediately neighboring nodes) will not be detected by the poisoned
reverse technique.
A Comparison of LS and DV Routing Algorithms
The DV and LS algorithms take complementary approaches towards computing
routing. In the DV algorithm, each node talks to only its directly connected neighbors, but it provides its neighbors with least-cost estimates from itself to all the
nodes (that it knows about) in the network. In the LS algorithm, each node talks with
all other nodes (via broadcast), but it tells them only the costs of its directly connected links. Let’s conclude our study of LS and DV algorithms with a quick comparison of some of their attributes. Recall that N is the set of nodes (routers) and E
is the set of edges (links).
• Message complexity. We have seen that LS requires each node to know the
cost of each link in the network. This requires O(|N| |E|) messages to be sent.
Also, whenever a link cost changes, the new link cost must be sent to all
nodes. The DV algorithm requires message exchanges between directly connected neighbors at each iteration. We have seen that the time needed for the
algorithm to converge can depend on many factors. When link costs change,
the DV algorithm will propagate the results of the changed link cost only if
the new link cost results in a changed least-cost path for one of the nodes
attached to that link.
• Speed of convergence. We have seen that our implementation of LS is an O(|N|2)
algorithm requiring O(|N| |E|)) messages. The DV algorithm can converge slowly
and can have routing loops while the algorithm is converging. DV also suffers
from the count-to-infinity problem.
• Robustness. What can happen if a router fails, misbehaves, or is sabotaged?
Under LS, a router could broadcast an incorrect cost for one of its attached
links (but no others). A node could also corrupt or drop any packets it received
as part of an LS broadcast. But an LS node is computing only its own forwarding tables; other nodes are performing similar calculations for themselves. This
means route calculations are somewhat separated under LS, providing a degree
of robustness. Under DV, a node can advertise incorrect least-cost paths to any
or all destinations. (Indeed, in 1997, a malfunctioning router in a small ISP
4.5
•
ROUTING ALGORITHMS
provided national backbone routers with erroneous routing information. This
caused other routers to flood the malfunctioning router with traffic and caused
large portions of the Internet to become disconnected for up to several hours
[Neumann 1997].) More generally, we note that, at each iteration, a node’s calculation in DV is passed on to its neighbor and then indirectly to its neighbor’s
neighbor on the next iteration. In this sense, an incorrect node calculation can
be diffused through the entire network under DV.
In the end, neither algorithm is an obvious winner over the other; indeed, both algorithms are used in the Internet.
Other Routing Algorithms
The LS and DV algorithms we have studied are not only widely used in practice,
they are essentially the only routing algorithms used in practice today in the Internet. Nonetheless, many routing algorithms have been proposed by researchers
over the past 30 years, ranging from the extremely simple to the very sophisticated
and complex. A broad class of routing algorithms is based on viewing packet traffic as flows between sources and destinations in a network. In this approach, the
routing problem can be formulated mathematically as a constrained optimization
problem known as a network flow problem [Bertsekas 1991]. Yet another set of
routing algorithms we mention here are those derived from the telephony world.
These circuit-switched routing algorithms are of interest to packet-switched
data networking in cases where per-link resources (for example, buffers, or a fraction of the link bandwidth) are to be reserved for each connection that is routed
over the link. While the formulation of the routing problem might appear quite
different from the least-cost routing formulation we have seen in this chapter,
there are a number of similarities, at least as far as the path-finding algorithm
(routing algorithm) is concerned. See [Ash 1998; Ross 1995; Girard 1990] for a
detailed discussion of this research area.
4.5.3 Hierarchical Routing
In our study of LS and DV algorithms, we’ve viewed the network simply as a collection of interconnected routers. One router was indistinguishable from another in
the sense that all routers executed the same routing algorithm to compute routing
paths through the entire network. In practice, this model and its view of a homogenous set of routers all executing the same routing algorithm is a bit simplistic for at
least two important reasons:
• Scale. As the number of routers becomes large, the overhead involved in
computing, storing, and communicating routing information (for example,
379
380
CHAPTER 4
•
THE NETWORK LAYER
LS updates or least-cost path changes) becomes prohibitive. Today’s public
Internet consists of hundreds of millions of hosts. Storing routing information at
each of these hosts would clearly require enormous amounts of memory. The
overhead required to broadcast LS updates among all of the routers in the public
Internet would leave no bandwidth left for sending data packets! A distance-vector algorithm that iterated among such a large number of routers would surely
never converge. Clearly, something must be done to reduce the complexity of
route computation in networks as large as the public Internet.
• Administrative autonomy. Although researchers tend to ignore issues such as a
company’s desire to run its routers as it pleases (for example, to run whatever
routing algorithm it chooses) or to hide aspects of its network’s internal organization from the outside, these are important considerations. Ideally, an organization should be able to run and administer its network as it wishes, while still
being able to connect its network to other outside networks.
Both of these problems can be solved by organizing routers into autonomous systems (ASs), with each AS consisting of a group of routers that are typically under
the same administrative control (e.g., operated by the same ISP or belonging to the
same company network). Routers within the same AS all run the same routing algorithm (for example, an LS or DV algorithm) and have information about each
other—exactly as was the case in our idealized model in the preceding section. The
routing algorithm running within an autonomous system is called an intraautonomous system routing protocol. It will be necessary, of course, to connect
ASs to each other, and thus one or more of the routers in an AS will have the added
task of being responsible for forwarding packets to destinations outside the AS;
these routers are called gateway routers.
Figure 4.32 provides a simple example with three ASs: AS1, AS2, and AS3.
In this figure, the heavy lines represent direct link connections between pairs of
routers. The thinner lines hanging from the routers represent subnets that are
directly connected to the routers. AS1 has four routers—1a, 1b, 1c, and 1d—
which run the intra-AS routing protocol used within AS1. Thus, each of these
four routers knows how to forward packets along the optimal path to any destination within AS1. Similarly, autonomous systems AS2 and AS3 each have three
routers. Note that the intra-AS routing protocols running in AS1, AS2, and AS3
need not be the same. Also note that the routers 1b, 1c, 2a, and 3a are all gateway
routers.
It should now be clear how the routers in an AS determine routing paths for
source-destination pairs that are internal to the AS. But there is still a big missing
piece to the end-to-end routing puzzle. How does a router, within some AS, know
how to route a packet to a destination that is outside the AS? It’s easy to answer
this question if the AS has only one gateway router that connects to only one
other AS. In this case, because the AS’s intra-AS routing algorithm has determined the least-cost path from each internal router to the gateway router, each
4.5
•
ROUTING ALGORITHMS
381
2c
3c
3a
AS3
2a
1c
1b
1a
AS2
3b
2b
1d
AS1
Intra-AS routing
algorithm
Inter-AS routing
algorithm
Forwarding
table
Figure 4.32
An example of interconnected autonomous systems
internal router knows how it should forward the packet. The gateway router, upon
receiving the packet, forwards the packet on the one link that leads outside the
AS. The AS on the other side of the link then takes over the responsibility
of routing the packet to its ultimate destination. As an example, suppose router
2b in Figure 4.32 receives a packet whose destination is outside of AS2. Router
2b will then forward the packet to either router 2a or 2c, as specified by router
2b’s forwarding table, which was configured by AS2’s intra-AS routing protocol.
The packet will eventually arrive to the gateway router 2a, which will forward
the packet to 1b. Once the packet has left 2a, AS2’s job is done with this one
packet.
So the problem is easy when the source AS has only one link that leads outside
the AS. But what if the source AS has two or more links (through two or more gateway routers) that lead outside the AS? Then the problem of knowing where to forward the packet becomes significantly more challenging. For example, consider a
router in AS1 and suppose it receives a packet whose destination is outside the AS.
The router should clearly forward the packet to one of its two gateway routers, 1b or
1c, but which one? To solve this problem, AS1 needs (1) to learn which destinations
are reachable via AS2 and which destinations are reachable via AS3, and (2) to
propagate this reachability information to all the routers within AS1, so that each
router can configure its forwarding table to handle external-AS destinations. These
382
CHAPTER 4
•
THE NETWORK LAYER
two tasks—obtaining reachability information from neighboring ASs and propagating the reachability information to all routers internal to the AS—are handled by the
inter-AS routing protocol. Since the inter-AS routing protocol involves communication between two ASs, the two communicating ASs must run the same inter-AS
routing protocol. In fact, in the Internet all ASs run the same inter-AS routing protocol, called BGP4, which is discussed in the next section. As shown in Figure 4.32,
each router receives information from an intra-AS routing protocol and an inter-AS
routing protocol, and uses the information from both protocols to configure its forwarding table.
As an example, consider a subnet x (identified by its CIDRized address), and
suppose that AS1 learns from the inter-AS routing protocol that subnet x is reachable from AS3 but is not reachable from AS2. AS1 then propagates this information
to all of its routers. When router 1d learns that subnet x is reachable from AS3, and
hence from gateway 1c, it then determines, from the information provided by the
intra-AS routing protocol, the router interface that is on the least-cost path from
router 1d to gateway router 1c. Say this is interface I. The router 1d can then put the
entry (x, I) into its forwarding table. (This example, and others presented in this section, gets the general ideas across but is a simplification of what really happens in
the Internet. In the next section we’ll provide a more detailed description, albeit
more complicated, when we discuss BGP.)
Following up on the previous example, now suppose that AS2 and AS3 connect to other ASs, which are not shown in the diagram. Also suppose that AS1
learns from the inter-AS routing protocol that subnet x is reachable both from AS2,
via gateway 1b, and from AS3, via gateway 1c. AS1 would then propagate this
information to all its routers, including router 1d. In order to configure its forwarding table, router 1d would have to determine to which gateway router, 1b or 1c, it
should direct packets that are destined for subnet x. One approach, which is often
employed in practice, is to use hot-potato routing. In hot-potato routing, the AS
gets rid of the packet (the hot potato) as quickly as possible (more precisely, as
inexpensively as possible). This is done by having a router send the packet to the
gateway router that has the smallest router-to-gateway cost among all gateways
with a path to the destination. In the context of the current example, hot-potato
routing, running in 1d, would use information from the intra-AS routing protocol
to determine the path costs to 1b and 1c, and then choose the path with the least
cost. Once this path is chosen, router 1d adds an entry for subnet x in its forwarding table. Figure 4.33 summarizes the actions taken at router 1d for adding the new
entry for x to the forwarding table.
When an AS learns about a destination from a neighboring AS, the AS can
advertise this routing information to some of its other neighboring ASs. For example,
suppose AS1 learns from AS2 that subnet x is reachable via AS2. AS1 could then tell
AS3 that x is reachable via AS1. In this manner, if AS3 needs to route a packet
destined to x, AS3 would forward the packet to AS1, which would in turn forward the
packet to AS2. As we’ll see in our discussion of BGP, an AS has quite a bit of
4.6
Learn from inter-AS
protocol that subnet
x is reachable via
multiple gateways.
Figure 4.33
Use routing info from
intra-AS protocol to
determine costs of
least-cost paths to
each of the gateways.
•
ROUTING IN THE INTERNET
Hot potato routing:
Choose the gateway
that has the
smallest least cost.
Steps in adding an outside-AS destination in a router’s forwarding table
flexibility in deciding which destinations it advertises to its neighboring ASs. This
is a policy decision, typically depending more on economic issues than on technical
issues.
Recall from Section 1.5 that the Internet consists of a hierarchy of interconnected ISPs. So what is the relationship between ISPs and ASs? You might think that
the routers in an ISP, and the links that interconnect them, constitute a single AS.
Although this is often the case, many ISPs partition their network into multiple ASs.
For example, some tier-1 ISPs use one AS for their entire network; others break up
their ISP into tens of interconnected ASs.
In summary, the problems of scale and administrative authority are solved by
defining autonomous systems. Within an AS, all routers run the same intra-AS routing protocol. Among themselves, the ASs run the same inter-AS routing protocol.
The problem of scale is solved because an intra-AS router need only know about
routers within its AS. The problem of administrative authority is solved since an
organization can run whatever intra-AS routing protocol it chooses; however, each
pair of connected ASs needs to run the same inter-AS routing protocol to exchange
reachability information.
In the following section, we’ll examine two intra-AS routing protocols (RIP and
OSPF) and the inter-AS routing protocol (BGP) that are used in today’s Internet.
These case studies will nicely round out our study of hierarchical routing.
4.6 Routing in the Internet
Having studied Internet addressing and the IP protocol, we now turn our attention to
the Internet’s routing protocols; their job is to determine the path taken by a datagram between source and destination. We’ll see that the Internet’s routing protocols
embody many of the principles we learned earlier in this chapter. The link-state and
distance-vector approaches studied in Sections 4.5.1 and 4.5.2 and the notion of an
autonomous system considered in Section 4.5.3 are all central to how routing is
done in today’s Internet.
383
Determine from
forwarding table the
interface I that leads
to least-cost gateway.
Enter (x,I) in
forwarding table.
384
CHAPTER 4
•
THE NETWORK LAYER
Recall from Section 4.5.3 that an autonomous system (AS) is a collection of
routers under the same administrative and technical control, and that all run the
same routing protocol among themselves. Each AS, in turn, typically contains multiple subnets (where we use the term subnet in the precise, addressing sense in Section 4.4.2).
4.6.1 Intra-AS Routing in the Internet: RIP
An intra-AS routing protocol is used to determine how routing is performed within
an autonomous system (AS). Intra-AS routing protocols are also known as interior
gateway protocols. Historically, two routing protocols have been used extensively
for routing within an autonomous system in the Internet: the Routing Information
Protocol (RIP) and Open Shortest Path First (OSPF). A routing protocol closely
related to OSPF is the IS-IS protocol [RFC 1142, Perlman 1999]. We first discuss
RIP and then consider OSPF.
RIP was one of the earliest intra-AS Internet routing protocols and is still in
widespread use today. It traces its origins and its name to the Xerox Network Systems (XNS) architecture. The widespread deployment of RIP was due in great part
to its inclusion in 1982 in the Berkeley Software Distribution (BSD) version of
UNIX supporting TCP/IP. RIP version 1 is defined in [RFC 1058], with a backwardcompatible version 2 defined in [RFC 2453].
RIP is a distance-vector protocol that operates in a manner very close to the
idealized DV protocol we examined in Section 4.5.2. The version of RIP specified
in RFC 1058 uses hop count as a cost metric; that is, each link has a cost of 1. In
the DV algorithm in Section 4.5.2, for simplicity, costs were defined between pairs
of routers. In RIP (and also in OSPF), costs are actually from source router to a destination subnet. RIP uses the term hop, which is the number of subnets traversed
along the shortest path from source router to destination subnet, including the destination subnet. Figure 4.34 illustrates an AS with six leaf subnets. The table in the
figure indicates the number of hops from the source A to each of the leaf subnets.
The maximum cost of a path is limited to 15, thus limiting the use of RIP to
autonomous systems that are fewer than 15 hops in diameter. Recall that in DV
protocols, neighboring routers exchange distance vectors with each other. The
distance vector for any one router is the current estimate of the shortest path
distances from that router to the subnets in the AS. In RIP, routing updates
are exchanged between neighbors approximately every 30 seconds using a
RIP response message. The response message sent by a router or host contains
a list of up to 25 destination subnets within the AS, as well as the sender’s
distance to each of those subnets. Response messages are also known as RIP
advertisements.
Let’s take a look at a simple example of how RIP advertisements work. Consider the portion of an AS shown in Figure 4.35. In this figure, lines connecting the
routers denote subnets. Only selected routers (A, B, C, and D) and subnets (w, x, y,
4.6
v
u
B
A
Hops
u
v
w
x
y
z
1
2
2
3
3
2
x
C
Figure 4.34
ROUTING IN THE INTERNET
Destination
w
D
y
z
•
Number of hops from source router A to various subnets
and z) are labeled. Dotted lines indicate that the AS continues on; thus this
autonomous system has many more routers and links than are shown.
Each router maintains a RIP table known as a routing table. A router’s routing
table includes both the router’s distance vector and the router’s forwarding table.
Figure 4.36 shows the routing table for router D. Note that the routing table has
three columns. The first column is for the destination subnet, the second column
indicates the identity of the next router along the shortest path to the destination subnet, and the third column indicates the number of hops (that is, the number of subnets that have to be traversed, including the destination subnet) to get to the
destination subnet along the shortest path. For this example, the table indicates that
to send a datagram from router D to destination subnet w, the datagram should first
be forwarded to neighboring router A; the table also indicates that destination subnet w is two hops away along the shortest path. Similarly, the table indicates that
subnet z is seven hops away via router B. In principle, a routing table will have one
row for each subnet in the AS, although RIP version 2 allows subnet entries to be
aggregated using route aggregation techniques similar to those we examined in
z
w
x
A
y
D
B
C
Figure 4.35
A portion of an autonomous system
385
386
CHAPTER 4
•
THE NETWORK LAYER
Destination Subnet
Next Router
Number of Hops to Destination
w
A
2
y
B
2
z
B
7
x
—
1
....
....
....
Figure 4.36
Routing table in router D before receiving advertisement
from router A
Section 4.4. The table in Figure 4.36, and the subsequent tables to come, are only
partially complete.
Now suppose that 30 seconds later, router D receives from router A the advertisement shown in Figure 4.37. Note that this advertisement is nothing other than
the routing table information from router A! This information indicates, in particular, that subnet z is only four hops away from router A. Router D, upon receiving this
advertisement, merges the advertisement (Figure 4.37) with the old routing table
(Figure 4.36). In particular, router D learns that there is now a path through router A
to subnet z that is shorter than the path through router B. Thus, router D updates its
routing table to account for the shorter shortest path, as shown in Figure 4.38. How
is it, you might ask, that the shortest path to subnet z has become shorter? Possibly,
the decentralized distance-vector algorithm is still in the process of converging (see
Section 4.5.2), or perhaps new links and/or routers were added to the AS, thus
changing the shortest paths in the AS.
Let’s next consider a few of the implementation aspects of RIP. Recall that
RIP routers exchange advertisements approximately every 30 seconds. If a router
does not hear from its neighbor at least once every 180 seconds, that neighbor is
considered to be no longer reachable; that is, either the neighbor has died or the
Destination Subnet
Next Router
Number of Hops to Destination
z
C
4
w
—
1
x
—
1
....
....
....
Figure 4.37
Advertisement from router A
4.6
•
ROUTING IN THE INTERNET
Destination Subnet
Next Router
Number of Hops to Destination
w
A
2
y
B
2
z
A
5
....
....
....
Figure 4.38
Routing table in router D after receiving advertisement from
router A
connecting link has gone down. When this happens, RIP modifies the local routing
table and then propagates this information by sending advertisements to its neighboring routers (the ones that are still reachable). A router can also request information about its neighbor’s cost to a given destination using RIP’s request message.
Routers send RIP request and response messages to each other over UDP using port
number 520. The UDP segment is carried between routers in a standard IP datagram. The fact that RIP uses a transport-layer protocol (UDP) on top of a networklayer protocol (IP) to implement network-layer functionality (a routing algorithm)
may seem rather convoluted (it is!). Looking a little deeper at how RIP is implemented will clear this up.
Figure 4.39 sketches how RIP is typically implemented in a UNIX system, for
example, a UNIX workstation serving as a router. A process called routed (pronounced
“route dee”) executes RIP, that is, maintains routing information and exchanges
messages with routed processes running in neighboring routers. Because RIP is
implemented as an application-layer process (albeit a very special one that is able to
Routed
Routed
Transport
(UDP)
Transport
(UDP)
Network
(IP)
Forwarding
tables
Forwarding
tables
Network
(IP)
Data link
Data link
Physical
Physical
Figure 4.39
Implementation of RIP as the routed daemon
387
388
CHAPTER 4
•
THE NETWORK LAYER
manipulate the routing tables within the UNIX kernel), it can send and receive messages over a standard socket and use a standard transport protocol. As shown, RIP is
implemented as an application-layer protocol (see Chapter 2) running over UDP. If
you’re interested in looking at an implementation of RIP (or the OSPF and BGP protocols that we will study shortly), see [Quagga 2012].
4.6.2 Intra-AS Routing in the Internet: OSPF
Like RIP, OSPF routing is widely used for intra-AS routing in the Internet. OSPF
and its closely related cousin, IS-IS, are typically deployed in upper-tier ISPs
whereas RIP is deployed in lower-tier ISPs and enterprise networks. The Open in
OSPF indicates that the routing protocol specification is publicly available (for
example, as opposed to Cisco’s EIGRP protocol). The most recent version of OSPF,
version 2, is defined in RFC 2328, a public document.
OSPF was conceived as the successor to RIP and as such has a number of
advanced features. At its heart, however, OSPF is a link-state protocol that uses
flooding of link-state information and a Dijkstra least-cost path algorithm. With
OSPF, a router constructs a complete topological map (that is, a graph) of the entire
autonomous system. The router then locally runs Dijkstra’s shortest-path algorithm
to determine a shortest-path tree to all subnets, with itself as the root node. Individual link costs are configured by the network administrator (see Principles and Practice: Setting OSPF Weights). The administrator might choose to set all link costs to
1, thus achieving minimum-hop routing, or might choose to set the link weights to
be inversely proportional to link capacity in order to discourage traffic from using
low-bandwidth links. OSPF does not mandate a policy for how link weights are set
(that is the job of the network administrator), but instead provides the mechanisms
(protocol) for determining least-cost path routing for the given set of link weights.
With OSPF, a router broadcasts routing information to all other routers in the
autonomous system, not just to its neighboring routers. A router broadcasts linkstate information whenever there is a change in a link’s state (for example, a change
in cost or a change in up/down status). It also broadcasts a link’s state periodically
(at least once every 30 minutes), even if the link’s state has not changed. RFC 2328
notes that “this periodic updating of link state advertisements adds robustness to the
link state algorithm.” OSPF advertisements are contained in OSPF messages that
are carried directly by IP, with an upper-layer protocol of 89 for OSPF. Thus, the
OSPF protocol must itself implement functionality such as reliable message transfer
and link-state broadcast. The OSPF protocol also checks that links are operational
(via a HELLO message that is sent to an attached neighbor) and allows an OSPF
router to obtain a neighboring router’s database of network-wide link state.
Some of the advances embodied in OSPF include the following:
• Security. Exchanges between OSPF routers (for example, link-state updates)
can be authenticated. With authentication, only trusted routers can participate
4.6
•
ROUTING IN THE INTERNET
in the OSPF protocol within an AS, thus preventing malicious intruders (or networking students taking their newfound knowledge out for a joyride) from
injecting incorrect information into router tables. By default, OSPF packets
between routers are not authenticated and could be forged. Two types of
authentication can be configured—simple and MD5 (see Chapter 8 for a discussion on MD5 and authentication in general). With simple authentication, the
same password is configured on each router. When a router sends an OSPF
packet, it includes the password in plaintext. Clearly, simple authentication is
not very secure. MD5 authentication is based on shared secret keys that are
configured in all the routers. For each OSPF packet that it sends, the router
computes the MD5 hash of the content of the OSPF packet appended with the
secret key. (See the discussion of message authentication codes in Chapter 7.)
Then the router includes the resulting hash value in the OSPF packet. The
receiving router, using the preconfigured secret key, will compute an MD5
hash of the packet and compare it with the hash value that the packet carries,
thus verifying the packet’s authenticity. Sequence numbers are also used with
MD5 authentication to protect against replay attacks.
• Multiple same-cost paths. When multiple paths to a destination have the same
cost, OSPF allows multiple paths to be used (that is, a single path need not be
chosen for carrying all traffic when multiple equal-cost paths exist).
• Integrated support for unicast and multicast routing. Multicast OSPF (MOSPF)
[RFC 1584] provides simple extensions to OSPF to provide for multicast routing
(a topic we cover in more depth in Section 4.7.2). MOSPF uses the existing
OSPF link database and adds a new type of link-state advertisement to the existing OSPF link-state broadcast mechanism.
• Support for hierarchy within a single routing domain. Perhaps the most significant advance in OSPF is the ability to structure an autonomous system hierarchically. Section 4.5.3 has already looked at the many advantages of hierarchical
routing structures. We cover the implementation of OSPF hierarchical routing in
the remainder of this section.
An OSPF autonomous system can be configured hierarchically into areas. Each
area runs its own OSPF link-state routing algorithm, with each router in an area
broadcasting its link state to all other routers in that area. Within each area, one or
more area border routers are responsible for routing packets outside the area. Lastly,
exactly one OSPF area in the AS is configured to be the backbone area. The primary
role of the backbone area is to route traffic between the other areas in the AS. The
backbone always contains all area border routers in the AS and may contain nonborder routers as well. Inter-area routing within the AS requires that the packet be first
routed to an area border router (intra-area routing), then routed through the backbone to the area border router that is in the destination area, and then routed to the
final destination.
389
390
CHAPTER 4
•
THE NETWORK LAYER
PRINCIPLES IN PRACTICE
SETTING OSPF LINK WEIGHTS
Our discussion of link-state routing has implicitly assumed that link weights are set, a routing algorithm such as OSPF is run, and traffic flows according to the routing tables computed by the LS algorithm. In terms of cause and effect, the link weights are given (i.e., they
come first) and result (via Dijkstra’s algorithm) in routing paths that minimize overall cost. In
this viewpoint, link weights reflect the cost of using a link (e.g., if link weights are inversely
proportional to capacity, then the use of high-capacity links would have smaller weights
and thus be more attractive from a routing standpoint) and Disjkstra’s algorithm serves to
minimize overall cost.
In practice, the cause and effect relationship between link weights and routing paths
may be reversed, with network operators configuring link weights in order to obtain routing
paths that achieve certain traffic engineering goals [Fortz 2000, Fortz 2002]. For example,
suppose a network operator has an estimate of traffic flow entering the network at each
ingress point and destined for each egress point. The operator may then want to put in
place a specific routing of ingress-to-egress flows that minimizes the maximum utilization
over all of the network’s links. But with a routing algorithm such as OSPF, the operator’s
main “knobs” for tuning the routing of flows through the network are the link weights. Thus,
in order to achieve the goal of minimizing the maximum link utilization, the operator must
find the set of link weights that achieves this goal. This is a reversal of the cause and effect
relationship—the desired routing of flows is known, and the OSPF link weights must be
found such that the OSPF routing algorithm results in this desired routing of flows.
OSPF is a relatively complex protocol, and our coverage here has been necessarily brief; [Huitema 1998; Moy 1998; RFC 2328] provide additional details.
4.6.3 Inter-AS Routing: BGP
We just learned how ISPs use RIP and OSPF to determine optimal paths for sourcedestination pairs that are internal to the same AS. Let’s now examine how paths are
determined for source-destination pairs that span multiple ASs. The Border Gateway Protocol version 4, specified in RFC 4271 (see also [RFC 4274), is the de facto
standard inter-AS routing protocol in today’s Internet. It is commonly referred to as
BGP4 or simply as BGP. As an inter-AS routing protocol (see Section 4.5.3), BGP
provides each AS a means to
1. Obtain subnet reachability information from neighboring ASs.
2. Propagate the reachability information to all routers internal to the AS.
3. Determine “good” routes to subnets based on the reachability information and
on AS policy.
4.6
•
ROUTING IN THE INTERNET
391
Most importantly, BGP allows each subnet to advertise its existence to the rest of
the Internet. A subnet screams “I exist and I am here,” and BGP makes sure that all
the ASs in the Internet know about the subnet and how to get there. If it weren’t for
BGP, each subnet would be isolated—alone and unknown by the rest of the Internet.
BGP Basics
BGP is extremely complex; entire books have been devoted to the subject and many
issues are still not well understood [Yannuzzi 2005]. Furthermore, even after having
read the books and RFCs, you may find it difficult to fully master BGP without having practiced BGP for many months (if not years) as a designer or administrator of
an upper-tier ISP. Nevertheless, because BGP is an absolutely critical protocol for
the Internet—in essence, it is the protocol that glues the whole thing together—we
need to acquire at least a rudimentary understanding of how it works. We begin by
describing how BGP might work in the context of the simple example network we
studied earlier in Figure 4.32. In this description, we build on our discussion of hierarchical routing in Section 4.5.3; we encourage you to review that material.
In BGP, pairs of routers exchange routing information over semipermanent
TCP connections using port 179. The semi-permanent TCP connections for the network in Figure 4.32 are shown in Figure 4.40. There is typically one such BGP TCP
connection for each link that directly connects two routers in two different ASs;
thus, in Figure 4.40, there is a TCP connection between gateway routers 3a and 1c
and another TCP connection between gateway routers 1b and 2a. There are also
semipermanent BGP TCP connections between routers within an AS. In particular,
Figure 4.40 displays a common configuration of one TCP connection for each pair
of routers internal to an AS, creating a mesh of TCP connections within each AS.
For each TCP connection, the two routers at the end of the connection are called
BGP peers, and the TCP connection along with all the BGP messages sent over the
2c
3c
3a
AS3
1c
2a
1a
1b
3b
AS2
Key:
eBGP session
iBGP session
Figure 4.40
VideoNote
Gluing the Internet
together
eBGP and iBGP sessions
1d
AS1
2b
392
CHAPTER 4
•
THE NETWORK LAYER
PRINCIPLES IN PRACTICE
OBTAINING INTERNET PRESENCE: PUTTING THE PUZZLE TOGETHER
Suppose you have just created a small that has a number of servers, including a public
Web server that describes your company’s products and services, a mail server from which
your employees obtain their email messages, and a DNS server. Naturally, you would like
the entire world to be able to surf your Web site in order to learn about your exciting products and services. Moreover, you would like your employees to be able to send and
receive email to potential customers throughout the world.
To meet these goals, you first need to obtain Internet connectivity, which is done by
contracting with, and connecting to, a local ISP. Your company will have a gateway
router, which will be connected to a router in your local ISP. This connection might be
a DSL connection through the existing telephone infrastructure, a leased line to the ISP’s
router, or one of the many other access solutions described in Chapter 1. Your local
ISP will also provide you with an IP address range, e.g., a /24 address range consisting of 256 addresses. Once you have your physical connectivity and your IP address
range, you will assign one of the IP addresses (in your address range) to your Web
server, one to your mail server, one to your DNS server, one to your gateway router,
and other IP addresses to other servers and networking devices in your company’s
network.
In addition to contracting with an ISP, you will also need to contract with an Internet registrar to obtain a domain name for your company, as described in Chapter 2. For example, if
your company’s name is, say, Xanadu Inc., you will naturally try to obtain the domain name
xanadu.com. Your company must also obtain presence in the DNS system. Specifically,
because outsiders will want to contact your DNS server to obtain the IP addresses of your
servers, you will also need to provide your registrar with the IP address of your DNS server.
Your registrar will then put an entry for your DNS server (domain name and corresponding IP
address) in the .com top-level-domain servers, as described in Chapter 2. After this step is
completed, any user who knows your domain name (e.g., xanadu.com) will be able to
obtain the IP address of your DNS server via the DNS system.
So that people can discover the IP addresses of your Web server, in your DNS server
you will need to include entries that map the host name of your Web server (e.g.,
www.xanadu.com) to its IP address. You will want to have similar entries for other publicly
available servers in your company, including your mail server. In this manner, if Alice
wants to browse your Web server, the DNS system will contact your DNS server, find the
IP address of your Web server, and give it to Alice. Alice can then establish a TCP
connection directly with your Web server.
However, there still remains one other necessary and crucial step to allow outsiders
from around the world access your Web server. Consider what happens when Alice,
who knows the IP address of your Web server, sends an IP datagram (e.g., a TCP SYN
segment) to that IP address. This datagram will be routed through the Internet, visiting a
series of routers in many different ASes, and eventually reach your Web server. When
4.6
•
ROUTING IN THE INTERNET
any one of the routers receives the datagram, it is going to look for an entry in its
forwarding table to determine on which outgoing port it should forward the datagram.
Therefore, each of the routers needs to know about the existence of your company’s
/24 prefix (or some aggregate entry). How does a router become aware of your
company’s prefix? As we have just seen, it becomes aware of it from BGP! Specifically,
when your company contracts with a local ISP and gets assigned a prefix (i.e., an
address range), your local ISP will use BGP to advertise this prefix to the ISPs to which
it connects. Those ISPs will then, in turn, use BGP to propagate the advertisement.
Eventually, all Internet routers will know about your prefix (or about some aggregate that
includes your prefix) and thus be able to appropriately forward datagrams destined to
your Web and mail servers.
connection is called a BGP session. Furthermore, a BGP session that spans two ASs
is called an external BGP (eBGP) session, and a BGP session between routers in
the same AS is called an internal BGP (iBGP) session. In Figure 4.40, the eBGP
sessions are shown with the long dashes; the iBGP sessions are shown with the short
dashes. Note that BGP session lines in Figure 4.40 do not always correspond to the
physical links in Figure 4.32.
BGP allows each AS to learn which destinations are reachable via its neighboring ASs. In BGP, destinations are not hosts but instead are CIDRized prefixes, with
each prefix representing a subnet or a collection of subnets. Thus, for example, suppose there are four subnets attached to AS2: 138.16.64/24, 138.16.65/24,
138.16.66/24, and 138.16.67/24. Then AS2 could aggregate the prefixes for these four
subnets and use BGP to advertise the single prefix to 138.16.64/22 to AS1. As another
example, suppose that only the first three of those four subnets are in AS2 and the
fourth subnet, 138.16.67/24, is in AS3. Then, as described in the Principles and Practice in Section 4.4.2, because routers use longest-prefix matching for forwarding datagrams, AS3 could advertise to AS1 the more specific prefix 138.16.67/24 and AS2
could still advertise to AS1 the aggregated prefix 138.16.64/22.
Let’s now examine how BGP would distribute prefix reachability information
over the BGP sessions shown in Figure 4.40. As you might expect, using the eBGP
session between the gateway routers 3a and 1c, AS3 sends AS1 the list of prefixes
that are reachable from AS3; and AS1 sends AS3 the list of prefixes that are reachable from AS1. Similarly, AS1 and AS2 exchange prefix reachability information
through their gateway routers 1b and 2a. Also as you may expect, when a gateway
router (in any AS) receives eBGP-learned prefixes, the gateway router uses its iBGP
sessions to distribute the prefixes to the other routers in the AS. Thus, all the routers
in AS1 learn about AS3 prefixes, including the gateway router 1b. The gateway
router 1b (in AS1) can therefore re-advertise AS3’s prefixes to AS2. When a router
(gateway or not) learns about a new prefix, it creates an entry for the prefix in its
forwarding table, as described in Section 4.5.3.
393
394
CHAPTER 4
•
THE NETWORK LAYER
Path Attributes and BGP Routes
Having now a preliminary understanding of BGP, let’s get a little deeper into it
(while still brushing some of the less important details under the rug!). In BGP, an
autonomous system is identified by its globally unique autonomous system number (ASN) [RFC 1930]. (Technically, not every AS has an ASN. In particular, a socalled stub AS that carries only traffic for which it is a source or destination will not
typically have an ASN; we ignore this technicality in our discussion in order to better see the forest for the trees.) AS numbers, like IP addresses, are assigned by
ICANN regional registries [ICANN 2012].
When a router advertises a prefix across a BGP session, it includes with the prefix a number of BGP attributes. In BGP jargon, a prefix along with its attributes is
called a route. Thus, BGP peers advertise routes to each other. Two of the more
important attributes are AS-PATH and NEXT-HOP:
• AS-PATH. This attribute contains the ASs through which the advertisement for the
prefix has passed. When a prefix is passed into an AS, the AS adds its ASN to the ASPATH attribute. For example, consider Figure 4.40 and suppose that prefix
138.16.64/24 is first advertised from AS2 to AS1; if AS1 then advertises the prefix to
AS3, AS-PATH would be AS2 AS1. Routers use the AS-PATH attribute to detect and
prevent looping advertisements; specifically, if a router sees that its AS is contained
in the path list, it will reject the advertisement. As we’ll soon discuss, routers also use
the AS-PATH attribute in choosing among multiple paths to the same prefix.
• Providing the critical link between the inter-AS and intra-AS routing protocols, the
NEXT-HOP attribute has a subtle but important use. The NEXT-HOP is the router
interface that begins the AS-PATH. To gain insight into this attribute, let’s again refer
to Figure 4.40. Consider what happens when the gateway router 3a in AS3 advertises
a route to gateway router 1c in AS1 using eBGP. The route includes the advertised
prefix, which we’ll call x, and an AS-PATH to the prefix. This advertisement also
includes the NEXT-HOP, which is the IP address of the router 3a interface that leads
to 1c. (Recall that a router has multiple IP addresses, one for each of its interfaces.)
Now consider what happens when router 1d learns about this route from iBGP. After
learning about this route to x, router 1d may want to forward packets to x along the
route, that is, router 1d may want to include the entry (x, l) in its forwarding table,
where l is its interface that begins the least-cost path from 1d towards the gateway
router 1c. To determine l, 1d provides the IP address in the NEXT-HOP attribute to
its intra-AS routing module. Note that the intra-AS routing algorithm has determined
the least-cost path to all subnets attached to the routers in AS1, including to the subnet for the link between 1c and 3a. From this least-cost path from 1d to the 1c-3a subnet, 1d determines its router interface l that begins this path and then adds the entry
(x, l) to its forwarding table. Whew! In summary, the NEXT-HOP attribute is used by
routers to properly configure their forwarding tables.
• Figure 4.41 illustrates another situation where the NEXT-HOP is needed. In this figure, AS1 and AS2 are connected by two peering links. A router in AS1 could learn
4.6
•
ROUTING IN THE INTERNET
AS2
Two peering
links between
AS2 and AS1
Router learns
about a
route to x
AS1
Router learns about
another route to x
Key:
Route advertisements
message for destination x
Figure 4.41
NEXT-HOP attributes in advertisements are used to determine which peering link to use
about two different routes to the same prefix x. These two routes could have the same
AS-PATH to x, but could have different NEXT-HOP values corresponding to the different peering links. Using the NEXT-HOP values and the intra-AS routing algorithm, the router can determine the cost of the path to each peering link, and then
apply hot-potato routing (see Section 4.5.3) to determine the appropriate interface.
BGP also includes attributes that allow routers to assign preference metrics to
the routes, and an attribute that indicates how the prefix was inserted into BGP at
the origin AS. For a full discussion of route attributes, see [Griffin 2012; Stewart
1999; Halabi 2000; Feamster 2004; RFC 4271].
When a gateway router receives a route advertisement, it uses its import policy
to decide whether to accept or filter the route and whether to set certain attributes
such as the router preference metrics. The import policy may filter a route because
the AS may not want to send traffic over one of the ASs in the route’s AS-PATH.
The gateway router may also filter a route because it already knows of a preferable
route to the same prefix.
BGP Route Selection
As described earlier in this section, BGP uses eBGP and iBGP to distribute routes
to all the routers within ASs. From this distribution, a router may learn about more
than one route to any one prefix, in which case the router must select one of the
395
396
CHAPTER 4
•
THE NETWORK LAYER
possible routes. The input into this route selection process is the set of all routes that
have been learned and accepted by the router. If there are two or more routes to the
same prefix, then BGP sequentially invokes the following elimination rules until one
route remains:
• Routes are assigned a local preference value as one of their attributes. The local
preference of a route could have been set by the router or could have been
learned by another router in the same AS. This is a policy decision that is left up
to the AS’s network administrator. (We will shortly discuss BGP policy issues in
some detail.) The routes with the highest local preference values are selected.
• From the remaining routes (all with the same local preference value), the route with
the shortest AS-PATH is selected. If this rule were the only rule for route selection,
then BGP would be using a DV algorithm for path determination, where the distance metric uses the number of AS hops rather than the number of router hops.
• From the remaining routes (all with the same local preference value and the same
AS-PATH length), the route with the closest NEXT-HOP router is selected. Here,
closest means the router for which the cost of the least-cost path, determined by
the intra-AS algorithm, is the smallest. As discussed in Section 4.5.3, this process
is called hot-potato routing.
• If more than one route still remains, the router uses BGP identifiers to select the
route; see [Stewart 1999].
The elimination rules are even more complicated than described above. To avoid
nightmares about BGP, it’s best to learn about BGP selection rules in small doses!
PRINCIPLES IN PRACTICE
PUTTING IT ALL TOGETHER: HOW DOES AN ENTRY GET INTO A ROUTER’S
FORWARDING TABLE?
Recall that an entry in a router’s forwarding table consists of a prefix (e.g., 138.16.64/22)
and a corresponding router output port (e.g., port 7). When a packet arrives to the router,
the packet’s destination IP address is compared with the prefixes in the forwarding table to
find the one with the longest prefix match. The packet is then forwarded (within the router)
to the router port associated with that prefix. Let’s now summarize how a routing entry
(prefix and associated port) gets entered into a forwarding table. This simple exercise will
tie together a lot of what we just learned about routing and forwarding. To make things
interesting, let’s assume that the prefix is a “foreign prefix,” that is, it does not belong to
the router’s AS but to some other AS.
In order for a prefix to get entered into the router’s forwarding table, the router has to
first become aware of the prefix (corresponding to a subnet or an aggregation of subnets). As we have just learned, the router becomes aware of the prefix via a BGP route
4.6
•
ROUTING IN THE INTERNET
advertisement. Such an advertisement may be sent to it over an eBGP session (from a
router in another AS) or over an iBGP session (from a router in the same AS).
After the router becomes aware of the prefix, it needs to determine the appropriate output
port to which datagrams destined to that prefix will be forwarded, before it can enter that
prefix in its forwarding table. If the router receives more than one route advertisement for this
prefix, the router uses the BGP route selection process, as described earlier in this subsection,
to find the “best” route for the prefix. Suppose such a best route has been selected. As
described earlier, the selected route includes a NEXT-HOP attribute, which is the IP address of
the first router outside the router’s AS along this best route. As described above, the router
then uses its intra-AS routing protocol (typically OSPF) to determine the shortest path to the
NEXT-HOP router. The router finally determines the port number to associate with the prefix
by identifying the first link along that shortest path. The router can then (finally!) enter the
prefix-port pair into its forwarding table! The forwarding table computed by the routing
processor (see Figure 4.6) is then pushed to the router’s input port line cards.
Routing Policy
Let’s illustrate some of the basic concepts of BGP routing policy with a simple example. Figure 4.42 shows six interconnected autonomous systems: A, B, C, W, X, and Y.
It is important to note that A, B, C, W, X, and Y are ASs, not routers. Let’s assume that
autonomous systems W, X, and Y are stub networks and that A, B, and C are backbone
provider networks. We’ll also assume that A, B, and C, all peer with each other, and
provide full BGP information to their customer networks. All traffic entering a stub
network must be destined for that network, and all traffic leaving a stub network must
have originated in that network. W and Y are clearly stub networks. X is a multihomed stub network, since it is connected to the rest of the network via two different
providers (a scenario that is becoming increasingly common in practice). However,
like W and Y, X itself must be the source/destination of all traffic leaving/entering X.
But how will this stub network behavior be implemented and enforced? How will X
be prevented from forwarding traffic between B and C? This can easily be
Key:
B
X
A
W
Customer
network
C
Y
Figure 4.42
A simple BGP scenario
Provider
network
397
398
CHAPTER 4
•
THE NETWORK LAYER
PRINCIPLES IN PRACTICE
WHY ARE THERE DIFFERENT INTER-AS AND INTRA-AS
ROUTING PROTOCOLS?
Having now studied the details of specific inter-AS and intra-AS routing protocols deployed
in today’s Internet, let’s conclude by considering perhaps the most fundamental question we
could ask about these protocols in the first place (hopefully, you have been wondering this
all along, and have not lost the forest for the trees!): Why are different inter-AS and intraAS routing protocols used?
The answer to this question gets at the heart of the differences between the goals of
routing within an AS and among ASs:
• Policy. Among ASs, policy issues dominate. It may well be important that traffic originating in a given AS not be able to pass through another specific AS. Similarly, a
given AS may well want to control what transit traffic it carries between other ASs. We
have seen that BGP carries path attributes and provides for controlled distribution of
routing information so that such policy-based routing decisions can be made. Within an
AS, everything is nominally under the same administrative control, and thus policy
issues play a much less important role in choosing routes within the AS.
• Scale. The ability of a routing algorithm and its data structures to scale to handle routing to/among large numbers of networks is a critical issue in inter-AS routing. Within
an AS, scalability is less of a concern. For one thing, if a single administrative domain
becomes too large, it is always possible to divide it into two ASs and perform inter-AS
routing between the two new ASs. (Recall that OSPF allows such a hierarchy to be
built by splitting an AS into areas.)
• Performance. Because inter-AS routing is so policy oriented, the quality (for example,
performance) of the routes used is often of secondary concern (that is, a longer or more
costly route that satisfies certain policy criteria may well be taken over a route that is
shorter but does not meet that criteria). Indeed, we saw that among ASs, there is not
even the notion of cost (other than AS hop count) associated with routes. Within a single AS, however, such policy concerns are of less importance, allowing routing to focus
more on the level of performance realized on a route.
accomplished by controlling the manner in which BGP routes are advertised. In particular, X will function as a stub network if it advertises (to its neighbors B and C) that
it has no paths to any other destinations except itself. That is, even though X may
know of a path, say XCY, that reaches network Y, it will not advertise this path to B.
Since B is unaware that X has a path to Y, B would never forward traffic destined to Y
(or C) via X. This simple example illustrates how a selective route advertisement policy can be used to implement customer/provider routing relationships.
4.7
•
BROADCAST AND MULTICAST ROUTING
Let’s next focus on a provider network, say AS B. Suppose that B has learned
(from A) that A has a path AW to W. B can thus install the route BAW into its routing information base. Clearly, B also wants to advertise the path BAW to its customer, X, so that X knows that it can route to W via B. But should B advertise the
path BAW to C? If it does so, then C could route traffic to W via CBAW. If A, B,
and C are all backbone providers, than B might rightly feel that it should not have
to shoulder the burden (and cost!) of carrying transit traffic between A and C. B
might rightly feel that it is A’s and C’s job (and cost!) to make sure that C can route
to/from A’s customers via a direct connection between A and C. There are currently
no official standards that govern how backbone ISPs route among themselves. However, a rule of thumb followed by commercial ISPs is that any traffic flowing across
an ISP’s backbone network must have either a source or a destination (or both) in a
network that is a customer of that ISP; otherwise the traffic would be getting a free
ride on the ISP’s network. Individual peering agreements (that would govern questions such as those raised above) are typically negotiated between pairs of ISPs and
are often confidential; [Huston 1999a] provides an interesting discussion of peering
agreements. For a detailed description of how routing policy reflects commercial
relationships among ISPs, see [Gao 2001; Dmitiropoulos 2007]. For a discussion of
BGP routing polices from an ISP standpoint, see [Caesar 2005b].
As noted above, BGP is the de facto standard for inter-AS routing for the public Internet. To see the contents of various BGP routing tables (large!) extracted
from routers in tier-1 ISPs, see http://www.routeviews.org. BGP routing tables
often contain tens of thousands of prefixes and corresponding attributes. Statistics
about the size and characteristics of BGP routing tables are presented in [Potaroo
2012].
This completes our brief introduction to BGP. Understanding BGP is important
because it plays a central role in the Internet. We encourage you to see the references
[Griffin 2012; Stewart 1999; Labovitz 1997; Halabi 2000; Huitema 1998; Gao
2001; Feamster 2004; Caesar 2005b; Li 2007] to learn more about BGP.
4.7 Broadcast and Multicast Routing
Thus far in this chapter, our focus has been on routing protocols that support unicast
(i.e., point-to-point) communication, in which a single source node sends a packet
to a single destination node. In this section, we turn our attention to broadcast and
multicast routing protocols. In broadcast routing, the network layer provides a
service of delivering a packet sent from a source node to all other nodes in the
network; multicast routing enables a single source node to send a copy of a packet
to a subset of the other network nodes. In Section 4.7.1 we’ll consider broadcast
routing algorithms and their embodiment in routing protocols. We’ll examine multicast routing in Section 4.7.2.
399
400
CHAPTER 4
•
THE NETWORK LAYER
4.7.1 Broadcast Routing Algorithms
Perhaps the most straightforward way to accomplish broadcast communication is
for the sending node to send a separate copy of the packet to each destination, as
shown in Figure 4.43(a). Given N destination nodes, the source node simply makes
N copies of the packet, addresses each copy to a different destination, and then
transmits the N copies to the N destinations using unicast routing. This N-wayunicast approach to broadcasting is simple—no new network-layer routing protocol, packet-duplication, or forwarding functionality is needed. There are, however,
several drawbacks to this approach. The first drawback is its inefficiency. If the
source node is connected to the rest of the network via a single link, then N separate
copies of the (same) packet will traverse this single link. It would clearly be more
efficient to send only a single copy of a packet over this first hop and then have the
node at the other end of the first hop make and forward any additional needed
copies. That is, it would be more efficient for the network nodes themselves (rather
than just the source node) to create duplicate copies of a packet. For example, in
Figure 4.43(b), only a single copy of a packet traverses the R1-R2 link. That packet
is then duplicated at R2, with a single copy being sent over links R2-R3 and R2-R4.
The additional drawbacks of N-way-unicast are perhaps more subtle, but no less
important. An implicit assumption of N-way-unicast is that broadcast recipients, and
their addresses, are known to the sender. But how is this information obtained? Most
likely, additional protocol mechanisms (such as a broadcast membership or
destination-registration protocol) would be required. This would add more overhead
and, importantly, additional complexity to a protocol that had initially seemed quite
simple. A final drawback of N-way-unicast relates to the purposes for which broadcast is to be used. In Section 4.5, we learned that link-state routing protocols use
broadcast to disseminate the link-state information that is used to compute unicast
routes. Clearly, in situations where broadcast is used to create and update unicast
routes, it would be unwise (at best!) to rely on the unicast routing infrastructure to
achieve broadcast.
Duplicate creation/transmission
R1
R1
Duplicate
R2
R3
a.
Figure 4.43
R2
R4
R3
Duplicate
R4
b.
Source-duplication versus in-network duplication
4.7
•
BROADCAST AND MULTICAST ROUTING
Given the several drawbacks of N-way-unicast broadcast, approaches in which
the network nodes themselves play an active role in packet duplication, packet forwarding, and computation of the broadcast routes are clearly of interest. We’ll
examine several such approaches below and again adopt the graph notation introduced in Section 4.5. We again model the network as a graph, G = (N,E), where N
is a set of nodes and a collection E of edges, where each edge is a pair of nodes from
N. We’ll be a bit sloppy with our notation and use N to refer to both the set of nodes,
as well as the cardinality (|N|) or size of that set when there is no confusion.
Uncontrolled Flooding
The most obvious technique for achieving broadcast is a flooding approach in
which the source node sends a copy of the packet to all of its neighbors. When a
node receives a broadcast packet, it duplicates the packet and forwards it to all of its
neighbors (except the neighbor from which it received the packet). Clearly, if the
graph is connected, this scheme will eventually deliver a copy of the broadcast
packet to all nodes in the graph. Although this scheme is simple and elegant, it has a
fatal flaw (before you read on, see if you can figure out this fatal flaw): If the graph
has cycles, then one or more copies of each broadcast packet will cycle indefinitely.
For example, in Figure 4.43, R2 will flood to R3, R3 will flood to R4, R4 will flood
to R2, and R2 will flood (again!) to R3, and so on. This simple scenario results in
the endless cycling of two broadcast packets, one clockwise, and one counterclockwise. But there can be an even more calamitous fatal flaw: When a node is connected to more than two other nodes, it will create and forward multiple copies of
the broadcast packet, each of which will create multiple copies of itself (at other
nodes with more than two neighbors), and so on. This broadcast storm, resulting
from the endless multiplication of broadcast packets, would eventually result in so
many broadcast packets being created that the network would be rendered useless.
(See the homework questions at the end of the chapter for a problem analyzing the
rate at which such a broadcast storm grows.)
Controlled Flooding
The key to avoiding a broadcast storm is for a node to judiciously choose when
to flood a packet and (e.g., if it has already received and flooded an earlier copy of
a packet) when not to flood a packet. In practice, this can be done in one of several
ways.
In sequence-number-controlled flooding, a source node puts its address (or
other unique identifier) as well as a broadcast sequence number into a broadcast
packet, then sends the packet to all of its neighbors. Each node maintains a list of
the source address and sequence number of each broadcast packet it has already
received, duplicated, and forwarded. When a node receives a broadcast packet, it
first checks whether the packet is in this list. If so, the packet is dropped; if not, the
401
402
CHAPTER 4
•
THE NETWORK LAYER
packet is duplicated and forwarded to all the node’s neighbors (except the node from
which the packet has just been received). The Gnutella protocol, discussed in Chapter 2, uses sequence-number-controlled flooding to broadcast queries in its overlay
network. (In Gnutella, message duplication and forwarding is performed at the
application layer rather than at the network layer.)
A second approach to controlled flooding is known as reverse path forwarding
(RPF) [Dalal 1978], also sometimes referred to as reverse path broadcast (RPB). The
idea behind RPF is simple, yet elegant. When a router receives a broadcast packet
with a given source address, it transmits the packet on all of its outgoing links (except
the one on which it was received) only if the packet arrived on the link that is on its
own shortest unicast path back to the source. Otherwise, the router simply discards
the incoming p