Storage Architectures for Server Blade Environments

Storage Architectures
for Server Blade
Environments
Abbott Schindler
Senior Technologist, HP
© 2004 Hewlett-Packard Development Company, L.P.
The information contained herein is subject to change without notice
Topics
•
Storage requirements driven by server blades
•
Migrating from traditional to blade environments
•
Storage infrastructure
•
Virtualization and provisioning
•
Deployment
August, 2004
Storage for Server Blades
2
Blade server adoption will drive new
storage deployments
Blades will account
for 25% of all
Server Shipments
by 2007
(IMEX Research)
August, 2004
Storage for Server Blades
3
The coming shifts for IT
infrastructures
•
•
•
All processes and content will be
transformed from physical and static
to digital, mobile and virtual.
The demand for simplicity,
manageability and adaptability will
change how customers work and
organize, buy and use technology.
There is going to be a shift…
− From hardware focus
− To ecosystem focus
•
The emerging focus:
− Managing the environment holistically
and provisioning it to the business
environment
August, 2004
Storage for Server Blades
4
Satisfying the shifts
Business
business processes
business metrics
and priorities,
SLAs
people & organizations
Management
IT processes
monitor and measure
control and change
SLAs, service
definition,
modeled
infrastructure
monitor and control interfaces
Infrastructure
virtualized
resources
systems
internet
August, 2004
storage
network
apps
services
internet
Storage for Server Blades
infrastructure
configuration
energy
internet
5
Key storage requirements
•
Requirements stem from blades moving toward
little embedded storage of their own
− Use embedded capacity for swap space, data caching,
and other temporary purposes
• Not OS boot image
• Not local data storage
•
Most compelling benefit of blades come when
connected to networked storage
− Have a ‘personality’ move from one blade to another
− Failover options
− Dynamic provisioning
August, 2004
Storage for Server Blades
6
Special considerations
•
HBAs go away
− GBICs, RNICs and other
networking connectivity is a
daughter card
− How to replace the “lost”
functionality?
•
Blade server racks will include
networking
− Ethernet switches
− Fibre channel switches
− FC and Ethernet connectivity
through the backplane
August, 2004
Storage for Server Blades
7
Blade server system overview
Built-in fabrics make blades an ideal
modular building block for virtualization
Blade systems unify
servers, storage, &
networking to increase
utilization and agility
Ultra-dense blades improve
data center utilization
Shared power infrastructure
reduces power consumption
Blades are optimized to
minimize installation
time and downtime
Blades are the bestmanaged industry
standard servers
August, 2004
Storage for Server Blades
Blades plug into a backplane
no cabling or cable management
Built in power and network redundancy
simplifies high availability
Standardized server, storage,
networking simplify management
Integrated lights-out management
provides secure access and control
from anywhere
8
Developing an overall storage
architecture
Advanced
Management
Dynamic
Provisioning
Basic
Management
Device
Config.
Interfaces
HBA
Information
Lifecycle Mgmt
Failover
Monitor/
Report
TOE
Security
RNIC
QoS-focused
management
Data
protection
Recovery
NAS Head/Filer
Virtualization
Network
Infrastructure
Fibre Channel
iSCSI
Tape Libraries
Optical
Ethernet
Storage Resources
Storage Arrays
August, 2004
Storage for Server Blades
NAS
9
The evolving environment
Today’s growing environment
connected clients
clustered
database
servers
deployment
server
development,
infrastructure and
business intelligence
servers
application
servers
O/S
Image
Disk
Tape
August, 2004
Storage for Server Blades
10
The evolving environment
Step 1 (or 2): consolidate storage
Standalone
NAS or
NAS/SAN
Fusion
Tape and
Optical
Libraries
MSA Family
Disk
Tape
EVA Family
August, 2004
O/S
Image
XP Family
Storage for Server Blades
11
The evolving environment
Step 2 (or 1): consolidate systems
Networked
Thin networked
clients
clients
scale out
blades with
cluster file
system
P-class
Web blade
servers
server
deployment
E-class blade
server
servers
scale up
approach with
traditional
servers
clustered
database
servers
August, 2004
Storage for Server Blades
12
The evolving environment
Step 3: stateless computing
Thin Clients
O/S
Image
Boot from SAN
Disk
Tape
August, 2004
Storage for Server Blades
13
HP delivers single pane of glass
management
Service Centric
Management
- Example: Service
linked together
Desk, Navigator
Enterprise Operations
Management
- HP OpenView examples:
Reporter, Network Node Manager
Storage Area Management
- Example: HP OpenView SAM
Element Management
- Common interface, HP StorageWorks
Command View, Systems Insight Manager
August, 2004
Storage for Server Blades
14
Providing storage to servers
•
Today: boot locally
PXE Server
− Configure storage
− Manually present LUN to blade
server
− Write any images needed by the
server to its LUNs
− Start the application
•
Boot Services
SAN
Boot from SAN
−
−
−
−
Configure storage
Present LUN to blade server
Boot blade to PXE server
Enable BIOS on server’s network
interface to see array
− Write any OS, images needed by
the server to its LUNs
− Add drivers to OS to see the storage
network
August, 2004
Storage for Server Blades
Blade Servers
Requires manual interaction with
three management interfaces
Server
Storage
Network
15
As blade connectivity to SANs
increases…
•
Extending the ‘reach’ of networked storage is
important
− More cost/connectivity options
• Ethernet SAN (iSCSI, RDMA, etc.)
− Lower cost storage systems
• Incorporate SATA, SAS devices
•
The number and types of servers supported
concurrently per controller port becomes
increasingly important
− Especially as PC blades are added to the mix
August, 2004
Storage for Server Blades
16
Fabrics within future data centers
•
•
•
•
•
•
August, 2004
Storage for Server Blades
server
resources
management systems
•
Data Center Fabrics
Fabrics connecting heterogeneous
“islands” of compute & storage
resources
“Ethernet everywhere” scaling across
the datacenter with RDMA/TCP
(iSER) efficiency
Routers & bridges translate between
heterogeneous infrastructure islands
Enabling Utility Computing
Static n-tier architecture (DISA)
enhanced by dynamic resource pools
Resource access managed by UDC,
ProLiant Essentials and HP OpenView
Functions virtualized over the fabric
•provisioning, monitoring, resource
management by policy, service-centric
•
•database
•application
•web
internet
fabric
infrastructure
storage
resources
•network switches
•load balancers
•firewalls
•routers/gateways
•SAN
•NAS
•switching
17
Data center network
infrastructure evolution
storage
Today
Tomorrow
• SAN
1, 2, Gigabit Fibre
Channel
• NAS (storage over IP)
networking
• 1 Gigabit Ethernet
• KVM over IP
• SAN
1, 2, 4, 8/10 Gb FC
• iSCSI (storage over
IP)
• 10 Gigabit
Ethernet
• IP acceleration
storage elements
data center
fabric
(TCP/IP and IPsec)
(Lights-out
management)
• IP Fabrics
(RDMA/TCP)
clustering
storage
fabric
fabric switches
compute
fabric
• Proprietary Solutions
(ServerNet, Myrinet, etc.)
• InfiniBand
Fabrics
compute elements
August, 2004
Storage for Server Blades
18
Fabric directions
Enhance IP infrastructure capabilities to increase solution
performance and enable platform extension while
leveraging Ethernet pervasiveness
Just better networking
• Consolidate ports
• Converge functions
• Centralize management
• Lower CPU utilization
• Free memory bandwidth
• Increased overall solution performance
Ethernet
remote
management
networking
storage
clustering
Network Processing
Simplify infrastructure
Improved Efficiency
NIC
TOE
Adapter
August, 2004
Storage for Server Blades
RNIC
CPU/Memory
19
As more information assets are
deployed
•
Ensuring cost-effective usage of all physical
assets is essential
− Automatic, dynamic provisioning
• Storage and servers
• To application requirements (SLA, QoS) rather than IT
requirements (generally, pure cost)
August, 2004
Storage for Server Blades
20
Automatic provisioning of storage
•
The ultimate goal
− Automatically configured storage
PXE Server
• Driven by QoS and other attributes
− Management Server recognizes
application workload requirement
Boot Services
• Management Server dynamically
presents appropriate LUNs to blade
•
SAN
Manual intervention is
significantly reduced
− Plug components in
− Set provisioning policies
− All done through a single
management interface
August, 2004
Storage for Server Blades
Storage
Provisioning
Storage Management
Applications
Blade Servers
21
Virtualization: blades and storage
‘scale in’ approach
Build on advantages of virtualized storage…
OS App
OS App
utilization
Content C
Database B
Content A
Vmware,
Microsoft Virtual Server
OS App
OS App
OS App
OS App
utilization
ProLiant DL servers
ProLiant BL blades
Virtualized storage + virtualized blade resources
• Excess capacity on smallest scale out building block
• Server maintenance measured in physical instantiation
• Partition at the OS, reduce application contention
•Leading choices: VMware, Microsoft Virtual Server
• Partition the Resource
•Leading choice (Windows): HP ProLiant Essentials Resource Partition
Manager
• Net effect: better utilization, efficiency, flexibility
August, 2004
Storage for Server Blades
22
Virtualization: blades and storage
scale-out architecture
Clustered File System
SAN / NAS
Cfs
OS
Cfs
OS
Cfs
OS
App B
Database B
Content A
App A
Leveraging virtual storage and clustered file systems
• Storage efficiencies: better storage utilization, more efficient admin usage
• Shared files (lock managed across nodes) with “clustered file systems”
• Leading choices: PolyServe, Sistina (RedHat)
• Parallel database clusters – scale-out databases (Oracle 9i RAC and 10g)
• Scale a large application across blades for higher performance and availability....
at lower cost than large SMP UNIX servers
August, 2004
Storage for Server Blades
23
Blades everywhere…
Extending to the desktop
•
Consolidated Client
Infrastructure (CCI)
− A total HP solution that enables
enterprise customers to
consolidate their desktop
infrastructures
− Provides a flexible, personalized
environment
− Supports static and dynamic
blade allocation
• Dynamic enables “less than one
blade” per user and is disaster
tolerant
• Static allocates 1 blade/user;
appropriate for users needing
system level access
August, 2004
Storage for Server Blades
External
users
Internal
users
Firewall
Load
balancer
DNE
Blade PC’s
Storage
virtualization
NAS/SAN pool
24
HP’s Consolidated Client Infrastructure
Flexible personalized environment
Today’s PC
CCI solution
Access Tier
HP thin client
TFT display
1:1 blade to thin client via
RDP + standard Ethernet
HP Blade PC bc1000
In the
Datacenter
Access, compute and
storage all in one
package
Compute Tier Proliant e-class
Enclosure / Switch
Load Balancer
SAN or NAS
Resource Tier Shared printers
App servers
August, 2004
Storage for Server Blades
25
StorageWorks Scalable File Share
simplification with a clustered file system
Storage Clients (CFS)
Workstations & Application Servers (trad. Servers,
blade and PC servers)
Attributes
• Highly scalable (performance,
capacity, # clients)
• Single system image
• Shared data access (includes lock
management)
Metadata Servers(MDS)
Metadata Access
Concurrency
Serve File Metadata
Parallel Data Transfer
File Locking
Storage Control
(recovery, file
status, file
creation)
Object Storage Targets (OST)
SAN
August, 2004
Serve File Data
Storage for Server Blades
26
Storage architecture for the future
StorageWorks Grid
High
performance
inter-smart cell
networks
Hostconnected
networks
Storage clients
August, 2004
Storage for Server Blades
27
Planning for the future
Capacity
Time
Plan for the future, deal
with today
Acquire with both
“today” and “tomorrow”
perspectives
SAN
Refine overall
management as feasible
HP StorageWorks Grid
August, 2004
Storage for Server Blades
28
HP Services for blades and storage
•
Total IT services from a single provider
•
IT consolidation services
•
Architecture & design consulting services
•
Storage infrastructure solutions
•
Migration services
•
Storage deployment services
•
Mission-critical support services
•
Support services
•
Multi-vendor environment services
•
Managed services
August, 2004
Storage for Server Blades
Customers are looking for a
knowledgeable, experienced
service provider to help with
their IT infrastructures
29
HP blades and network storage together
deliver more for the Adaptive Enterprise
HP
Blade Servers
HP
StorageWorks
HP Management Tools
HP Services
More
August, 2004
Storage for Server Blades
Value
Simplicity
Agility
30
Customer example:
KT FreeTel Co. Ltd. (KTF)
Customer challenge
•
•
KTF – large Telco and
Network Service Provider
Goals
– Telecom Billing Solution
for IP data billing
– Improve TCO of server
infrastructure
– Reduce cost of SUN
infrastructure (installed
base)
August, 2004
HP solution
Customer results
Infrastructure
• 140 ProLiant BL blade server
farm: 2p p-class blades
servers running Linux
• HP StorageWorks EVA SAN
(3TB of storage)
Applications
• Applications developed by
solution partners for blade
server infrastructure
• IP billing and usage collection
application.
Lower cost of infrastructure –
in comparison to Sun
Microsystems
Why HP?
• Excellent account team
partnership
• HP products and Services
Storage for Server Blades
•
31
Customer example:
Large U.S. bank/financial service
Migration from Solaris SMP to Linux Blade cluster
Customer challenge
•
•
Large NA Bank – a
transaction reconciliation
application
Goals
– Dramatically improve the
price/performance of
server platform
– Ability to manage the
blade cluster as one
machine
– Successfully migrate a
Solaris application to
Linux
August, 2004
HP solution
Customer results
Infrastructure
• ProLiant BL blade server farm:
2p p-class blades servers
running Linux
• HP Blade management tools:
RDP, Insight Manger
• StorageWorks EVA SAN
Applications
• Clustered file system from
PolyServe
• F5 Networks Big-IP Blade
Controller S/W for virtual IP
load balancing of the cluster
• Custom application migrated
from Solaris
Application has improved
performance by 25%
• Platform costs reduced by
70%
Why HP?
• Good account team
relationship
• Demonstrated technical
expertise in implementing HP
ProLiant blades and SAN
connectivity
• Blade Management tools
• Long term investment
protection of blade system
infrastructure
Storage for Server Blades
•
32
HP blades and network storage together
deliver more for the Adaptive Enterprise
•
Consistent hardware implementations across the data center
•
Remove 1 to 1 relationship of hardware and image
Simplicity
−Boot servers from SAN for quick and consistent O/S and app image
−Server image or client “personality” can reside on any blade
−Users can access their desktop personality from anywhere
•
Centralized Back-up via SAN arrays and libraries
•
Quickly deploy and change data center
–Quickly present applications with Blade/SAN environment
–Quickly present application data to any blade (host)
Agility
–Rapidly change server role to meet business requirements in real-time
•
Architectural approach for growth
–SAN storage enables online volume growth beyond local disk capacities
–Virtual blade/storage enables utility-like computing
Value
August, 2004
•
Consolidation to increase utilization and reduce costs
•
Network storage enables high availability for maximum uptime
−Clustering
−Remote Replication
Storage for Server Blades
33
Summary
•
Modular storage has been “bladed” for several years; for
storage, it was needed to provide the sort of agility,
extensibility, and flexibility demanded for complex server
environments. Blades are really an extension of those
environments.
•
The best storage infrastructure for blade environments —
for both PC blades and server blades — is networked
storage.
− Server blades: either FC- or Ethernet-based SAN
− CCI: NAS, NAS/SAN fusion
•
Server consolidation (to blades) and storage consolidation
(to networked storage) go hand-in-hand.
August, 2004
Storage for Server Blades
34
August, 2004
Storage for Server Blades
35