Introduction to Scaling Networks

Introduction to Scaling
Networks
Routing and Switching: Scaling Networks – Chapter 1
Copyleft 2014 Hacklab Cosenza (http://hlcs.it)
Released under Creative Commons License 3.0 By-Sa
Cisco name, logo and materials are Copyright Cisco Systems Inc. 1
Scaling Computer Networks
●
●
●
●
●
Businesses can grow, along with their needs in terms of
users, uptime, speed. Their nework must be able to do the same.
A computer network designed to accomodate the need for growth
is called scalable.
Enterprise networks needs enterprise hardware, specifically
designed to provide reliability through redundancy and failover.
But nothing can replace a properly designed network.
First need of a proper design: the traffic must be organized so
that it doesn't unnecessarily propagate through the network at
the cost of bandwitdth.
We achieve this through a hierarchical design composed of 3
layers, bottom-to-top: Access, Distribution, Core. Each layer ha
s a specific function.
Hierarchical Network Design
●
●
●
●
●
The Access Layer is in charge of providing a connection to
the final users in a local network.
The Distribution Layer forwards traffic from/to different local
networks.
The Core Layer is a high speed backbone between
distruibution blocks, that is networks that are geographically
dispersed.
Crucially, user traffic starts at the access layer and it's not
passed to the upper layer unless their functionality is
required.
In smaller enteprise networks, Distribution and Core Layers
can be collapsed into a single one, reducing cost and
complexity.
Cisco Enterprise Architecture
●
●
●
●
●
●
Cisco advices this network design principle that divides
networks into modular components, maintaining the 3 layers.
Its modules include: Enterprise Campus, Enterprise Edge,
Service Provider Edge, Remote.
The Enteprise Campus includes Access, Distribution and
Core Layers. The entire enterprise infrastructure belongs to this
module.
In the Access Layer, L2/L3 switches provides port density and
links/trunks to the distribution layer of the building.
Distribution layer aggregates these links/trunks with L3
devices that provides routing, access control and QoS.
The Core Layer provides very fast connectivity between
distribution layer modules and the Enterprise Edge modules.
Cisco Enterprise Architecture
●
●
●
●
●
Enterprise Campus can have additional submodules, such as
Server Farm module, Data Center module, Service Module.
These submodules provided network monitoring and unified
control of services such as wireless controlling or telephony
services.
The Enterprise Edge includes modules that connect the
enterprise to remote resources through the service provider
network: Internet connectivity, VPN tunnels and WAN.
The Service Provider Edge provides services through a
service provider such as internet connection and phone lines.
Each of these services has a corrisponding edge device in
the Enterprise Edge that examines all packet and decides
whether it should be allowed on the enteprise network.
Failure Domains
●
●
●
●
●
A failure domain is the area of the network that is impacted by a
problem or failure.
Much of the impact of a failure domain is determined by the
functionality provided by the faulty device.
When a problem occurs, smaller failure domains reduce the
impact of the problem, simplify troubleshooting and thus
decrease downtime.
Reducing failure domains in the core layer is expensive, so
designer concentrate on prevention. Making sure network
errors affect a smaller area in the distribution layer is easier.
Distribution device (L3 switches or routers) are usually deployed
in pair, with access devices evenly distributed. This is called
a switch block. Switch blocks act indipendently of the others.
Failure Domains
Tips for Scalability
●
●
●
●
●
●
Use modular devices than can easily expands their capabilities.
Some device are more easily deployable in clusters that simplify
management and configuration.
Use these modular devices in a 3-layer network design.
Think about an IPv4 or IPv6 addressing strategy that will
accomodate future users. Re-addressing later could be a major
disruption.
Use routers and L3 switches to limit broadcast and filter traffic to
the core.
Use redundant links between network devices. Link aggregation
(such as Cisco's EtherChannel) methods increase bandwidth
without more expensive media channels.
A scalable routing protocol that minimizes the size of the routing
table.
Tips for Redundancy
●
●
●
●
●
Redundancy is all about minimizing the chance of a single
point of failure.
One easy method to achieve redundancy is by installing
duplicate equipment, at least for critical devices.
Redundancy also means redundant paths that traverses the
network.
Redundancy in switches could cause traffic loops at L2.
Spanning Tree Protocol (STP) is a mechanism that prevents
them.
STP disables redundant paths until those paths are
necessary, for instance in case of failure.
EtherChannel
●
●
●
●
●
Multiple traffic links between access and distribution switches will
need to converge in a single outgoing link that could become
a bottleneck.
It is possible to tie multiple physical links together in a single
logical one, increasing the available bandwidth. This is called
link aggregation.
EtherChannel is a Cisco protocol that use ports of the existing
switches to create a logical EtherChannel interface.
Every configuration is done on the EtherChannel interface,
ensuring consistency between ports.
EtherChannel can do load balancing between the links that
are part of the EtherChannel interface, using several methods.
Wireless Connectivity
●
●
●
Wireless connectivity is crucial in expanding a network,
because it increases flexibility without excessive costs.
Implementing wireless connectivity require at the very minimum
●
A wireless NIC for end devices
●
An access point (AP) or wireless router
Wireless connectivity needs careful planning. Several things
must be considered, among them:
●
Types of devices
●
Coverage
●
Interference
●
Security
●
Speed and protocols
Tuning Routing Protocols
●
●
●
●
OSPF supports a two-layer hierarchical design, or multiarea
OSPF
All OSPF networks begin with Area 0, also called the backbone
area.
As the network is expanded, other, non-backbone areas can be
created.
All non-backbone areas must directly connect to area 0.
Switch Platforms
●
●
●
●
Cisco has 5 categories for enterprise switches: Campus LAN
switches, Cloud-Managed switches, Data Center switches,
Service Provider switches, and Virtual Networking switches.
Enterprise switches comes in several form factor:
●
Fixed or Modular configuration
●
Stackable or Non-Stackable
For switches that are mounted in racks, thickness is important. It
is measured in rack units (U).
These are the most commons parameters that need to be
evaluated when choosing a switch solution: cost, port density,
power specifications, reliability, port speed, frame buffers,
scalability.
Port Density
●
●
●
●
Port density is the number of port available on a single switch.
High-port density switches allow better use of space and
power. Consider a single 48-ports vs dual 24-ports switches,
than scale the advantages to networks made of 1000+ end
devices.
Modular switches can support extremely high port density
through multiple expansion cards. Catalyst 6500 switches can
support more than 1000 ports.
Link aggregation reduces available ports, something less of
an impact with modular switches.
Forwarding Rates
●
●
●
●
●
Forwarding Rate means how much data a switch is able to
process per second.
Wire Speed is the data rate a single port on the switch is able to
attain. It can be 100Mb/s, 1Gb/s, 10Gb/s or even more.
If forwarding rates are too low, switches cannot reach full wire
speed on all of their ports at the same time.
Access layer switches are limited by their uplink speed to
the distribution layer anyway, so they don't need to have high
forwarding rates, reducing costs.
At the distribution and core layers an high forwarding rates has
a much greater impact on overall network performance.
PoE – Power over Ethernet
●
●
●
●
PoE can provide the power a device needs to operate on the
same Ethernet cabling used for data transmission.
PoE allows flexibility because it means a device can be
installed any place reached by an Ethernet cable.
PoE-capable switched are much more expensive, so it needs
to be evaluated carefully if PoE is needed or not.
Some Cisco switches also support PoE pass-through, which
allows powering the devices in a cascading fashion starting
for an upstream power source.
Multilayer Switching
●
●
●
Multi-layer switches have the ability of building a routing
table, support several routing protocol, and forward IP packets
nearly at the same L2 forwarding speed.
They are able to do that thanks to specialized ASIC
processors and routines.
The trend is clear: the cost of supporting routing in switches
is decreasing. Soon all switches will support it.
Router Requirements
●
●
●
Routing is an essential component of the distrubution layer. Without a
routing process packets cannot leave the LAN.
Routers provides several functions:
●
Selection of the appropriate path for the traffic.
●
Translation between different media types/protocols.
●
Broadcast containment
●
Connect remote location
●
Group users logically
●
Provides enhanced security
There are 3 categories of Cisco routers:
●
●
●
Branch Routers (for local routing and WAN access),
Network Edge Routers (boundary routers designed for multi-WAN
deployment),
Service Provider Routers (designed for Service providers who wants to
differentiate their services).
Router Hardware
●
●
Routers can have many form factors, from small desktop router
to rack or blade models.
Routers can be:
●
●
Fixed: the desired router interfaces are built-in and not
expandible.
Modular: the router has slots through which it's possible to
change the interfaces or adding new ones.
Managing IOS Files and Licenses
●
●
●
●
Cisco has a single operating system, IOS, for most of their
products.
IOS comes in many versions, and for a single version it has
many different images with different feature sets.
This feature sets are customizable by the user when a
product is purchased.
A new device is shipped with an image preinstalled that
corresponds to the customer-specified packages, and with a
permanent license for it.
In-Band vs Out-of-Band
●
●
●
●
●
There are two methods for connecting to a Cisco IOS device for
configuring it or monitoring it: In-Band and Out-of-Band.
Out-of-Band is used for initial configuration and every time a
network connection becomes unavailable.
It requires direct connection to the console port as well as a
terminal emulation client.
In-Band management is used for normal maintainance and
configuration, and it works over a working network connection.
It requires at least one network interface active, connected and
operational, as well as SSH, telnet or HTTP access to the device.
End of lesson