You are on page 1of 18

®

Introducing WebSphere eXtreme Scale


March, 2010

Ted Kirby
WebSphere eXtreme Scale Technical Evangelist

© 2010 IBM Corporation


IBM Software Group | WebSphere software

Typical Application Architecture


Where’s the bottleneck in this picture?

© 2006 IBM Corporation 2


IBM Software Group | WebSphere software

Memory Hierarchy
Data must get to CPU local memory to be processed.
machine machine

CPU Memory 2ms CPU Memory


20ns network 20ns

20ms

Get Data From Access Time Factor

Memory 20 ns 20 x 10-9 sec 1

Disk
Over network 2 ms 2 x 10-3 sec 100,000

Disk 20 ms 20 x 10-3 sec 1,000,000

Net: Put data in memory. Exploit network for capacity and availability.

© 2006 IBM Corporation 3


IBM Software Group | WebSphere software

Cache
Keep a copy of high-use data in a chunk of memory called a cache.

Event Time Get Data From


Memory 2ms
Cache hit 20 ns Memory CPU
20ns Cache
Remote Cache hit 2 ms Over network network

Cache miss* 20 ms Disk

20ms
*On cache miss, put data from disk in cache,
hoping for cache hit next time.
Disk

+ Larger cache increases cache hits. DB


EIS
+ Cache hits reduce response time and load on the disk.

© 2006 IBM Corporation 4


IBM Software Group | WebSphere software

Traditional Cache eXtreme Scale Cache


machine machine
JVM JVM
JVM
Cache
App App Cache App App
A, B, C
Server Logic A Server Logic

JVM
JVM Cache
JVM
D, E, F
App App
App App Cache
Server Logic
Server Logic A

machine machine
JVM JVM
Cache Cache
eXtreme Scale data grid
JVM JVM
1,000s
Cache Cache
WAY bigger cache

© 2006 IBM Corporation 5


IBM Software Group | WebSphere software

Partitioning: The key to Linear Scaling


Encyclopedia is partitioned
into volumes on shelves.
A B C D E F G
eXtreme Scale partitions
caches into shards in A-Z
containers, which are stored
in JVMs. H I J K L M N

•A container holds a variable number of shards,


limited by JVM size (2GB, or 64-bit…) O P Q R S T

•Can have multiple containers on a machine.


•Elastic: The number of machines may vary
U V W X Y Z
over time. (Like opening more registers at the
supermarket.)
•Add machines to handle more load
•Automatic recovery from node failures
© 2006 IBM Corporation 6
IBM Software Group | WebSphere software

What is WebSphere eXtreme Scale?


A flexible framework for realizing high
performance, scalable and data-intensive New York San Francisco
applications

London Shanghai

3 Use Cases
Platform for building
Form of in-memory database powerful XTP/Data Grid
 Manage application state applications
Very powerful cache
 Scales from simple in-  Scales to 1000’s of servers
process topologies to  Sometimes referred to as
powerful distributed Distributed Application
topologies. State Management.

© 2006 IBM Corporation 7


IBM Software Group | WebSphere software

IBM WebSphere eXtreme Scale


 Proven mature product:
– Third major release of product with V7.0
– Public References
– Private References
– Used at some of the largest web sites in the
world

 Lightweight runtime footprint (15MB jar)

 Integrates with all versions of WebSphere and


almost any Java-based application container or
Java Virtual Machine (1.4.2 and higher)

 Proven multi-data center capabilities

 Proven low-latency access to data

© 2006 IBM Corporation 8 8


9
IBM Software Group | WebSphere software

Common set of patterns


 Side Cache
– Hibernate / OpenJPA L2 cache
App
App
– Dynamic Cache (ND 6.1/ ND 7.0) App
App
• Page fragment cache
• Web services cache
• Command cache
– SOA State store
– SOA Result cache
• ESB cache mediation
 Benefits
– Improved response time
– Reduced load on DB

© 2006 IBM Corporation 9 9


1
0
IBM Software Group | WebSphere software

Common set of patterns


 Inline backend cache
– Loaders used to integrate with an
existing data service
App
App
– Read through cache App
App
– Write through cache

 System of Record Data Store


– Cache is used as the system of
Loader
record
– Write behind technology pushes
changes asynchronously to the
backend.

© 2006 IBM Corporation 10


10
1
1
IBM Software Group | WebSphere software

Common set of patterns


 HTTP Session Store
– Why use database for short-lived data?
– Application uses standard HTTPSession APIs
– Can be shared between applications sharing same domain name
– Grid runs inside web cluster JVMs using affinity

 Application Session
– Application uses eXtreme Scale APIs to manipulate the shared session
– Works even when web apps use different domain names

 Multi data center support for sessions


– SOA Sandbox exercise coming soon

© 2006 IBM Corporation 11


11
1
2
IBM Software Group | WebSphere software

Common set of patterns


 XTP (eXtreme Transaction Processing)
– Lowest possible latency
– Application code runs in grid itself
• Map/Reduce API supported
– Events routed to correct partitions for processing
• Gateway forwards events to correct partition
• JVM hosting a partition directly subscribes to events
– Write behind
• Databases relegated to durable log and reports

© 2006 IBM Corporation 12


12
IBM Software Group | WebSphere software

Simple Caching Scenarios


Challenges
• Redundant calls, doing something over Benefits
and over again, on expensive systems • Free up expensive backend
• Generally, to access data that does not Offload Redundant systems for critical tasks
change much (e.g., user profiles) • Reduce costs of system cycles for
• Often, multiple systems accessed (e.g. to Processing
repetitive data retrieval
fetch different parts of user profile) • Increase performance through in-
• Takes too long to retrieve the data, which memory, network cache
gets worse as transaction volumes grow

Benefits
• Automatic “drop-in” eXtreme
Challenges
Scale cache without invasive
• Web sites that need better management HTTP Session coding changes
and automatic fail-over of Web sessions – Replication • Higher availability and
usually Commerce or retail-related sites
performance for revenue-
producing applications

Benefits
• Better performance: turbo-charge
Challenges WAS caching layer via eXtreme
• Web applications that use (WAS) Dynamic “Drop-In” extension Scale “Drop-In” cache with no
Cache service and need better for Dynamic Cache coding changes
performance and scalability of their • Better scalability: Entire eXtreme
caching investment
Service Provider Scale grid is now your WAS cache
• More efficient: Single seamless
network cache to store data

© 2006 IBM Corporation 13


IBM Software Group | WebSphere software

Client Usage: One of the largest retail US banks


Retail Banking & Investments Side Cache - ESB Cache Mediation
Next-generation Online Banking
22 Million –
Before: 700ms to login with 2 backend calls
online banking users – After: 20ms to login with profile cache access

$6M/year cost savings in MIPs reduction
35x $500k 20x –
700k logins per hour across 3 data centers
reduced reduced reduction – 8Gb of data transfer per hour between DC’s
response costs per in “lost –
60 million page views/day
times month sessions”
– 10 million logins per day
– 3 active data centers

© 2006 IBM Corporation 14


14
IBM Software Group | WebSphere software

Client Usage: Social Media Website


Inline system-of-record cache,
Entertainment with write-behind.
10x Partitioning smashes scaling barrier.
reduced
7 Billion response
requests per day times
Fantasy Sports Web Infrastructure
– Before: 60ms response time against database

After: improved to 6ms response time

450k concurrent users
– 80k requests per second up to 1M in 2011

6 weeks from concept to production

Grid ORM
App Grid
Grid
ORM
ORM Oracle

© 2006 IBM Corporation 15


15
IBM Software Group | WebSphere software

Benefit Summary

 Caching
–Improved response time
–Reduced load on DB

 Partitioning
–Smashes the scaling barrier
–Allows elasticity for linear scaling, predictable response
times

© 2006 IBM Corporation 16


IBM Software Group | WebSphere software

Next Steps
The Getting Started with WebSphere eXtreme Scale Wiki
http://bit.ly/eXtremeScaleStart -- First page is an annotated bibliography

“Understanding WebSphere eXtreme Scale and how it works” developerWorks article


http://bit.ly/eXtremeScaleIntro

Download WebSphere eXtreme Scale for free and build a trial app
http://bit.ly/eXtremeScaleTrial

Run WebSphere eXtreme Scale in the clouds! -- on Amazon EC2


http://bit.ly/extremescaleAMI

Engage with the Emerging Technology Community on My developerWorks


http://bit.ly/dwEmerge

Engage with the Emerging Technology Community on My developerWorks


http://facebook.com/websphereemerge

Read Billy Newport’s blog on XTP


http://www.devwebsphere.com/

© 2006 IBM Corporation 17


1
8
IBM Software Group | WebSphere software

© 2006 IBM Corporation 18

You might also like