Take Your Infrastructure To The Next Level Of Agility And Cost Savings–Dynamic Application Infrastructure

Take Your Infrastructure To The Next Level Of Agility And Cost Savings With A Dynamic

Application Infrastructure

06 Take Your Infrastructure to the Next Level of Agility and Cost Savings 2

The WebSphere Application Server Family –Special Purpose Capabilities

WebSphere Virtual Enterprise

WebSphere eXtreme Scale

WebSphere Compute Grid

World’s first Application Server private cloud

Manage performance, health, and editions of applications for non-stop operation

In-memory data grid for extreme transaction processing

Execute and manage Java batch processes

WebSphere Application Server Hypervisor Edition


SOF Needs A Flexible Infrastructure To Improve Service Quality While Reducing Costs

I need to make sure my applications meet

response time requirements and

ensure our top customers receive the

best service!

Data CenterManager

CIO

My infrastructure and labor costs keep going

up. How can I meet my requirements

without spending more?


Server Sprawl Is Out Of Control And Has Costly Consequences

� Low server resource utilization

� Excessive energy usage and heating problems

� Too many software licenses

� Additional staff required for platform management

� Unexplained outages

� Too much complexity

� Response to business requirements is too slow


Dedicated Servers Are Underutilized

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Work

load

Provision capacityfor peak workload

Idle Resource

Idle Resource

Provision for expected growth

Average utilization

Idle Resource

The old paradigm of one server dedicated to one application wastes your money.


Server Sprawl Drives Increasing Annual Operating Costs

Worldwide IT Spending on Servers, Power, Coolingand Management/Administration

$0

$50

$100

$150

$200

$250

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

0

5

10

15

20

25

30

35

40

45

50

Power and Cooling CostsServer Management and Admin CostsNew Server Spending

PhysicalServer InstalledBase (Millions)

SpendingUS$(B)

Source: IDC, 2008.


Server Virtualization And WebSphere Can Improve Utilization And Efficiency

� Server Virtualization with PowerVM� Increases server utilization reducing hardware, software, energy

and administration costs

� WebSphere Virtual Enterprise� Performance, health and application edition management for non-

stop operation

� Increases server utilization reducing hardware, software, energyand administration costs

� WebSphere Cloudburst� Improves labor productivity reducing administration costs


Virtualization And Consolidation On IBM Power Systems

Application

Linux

Application

AIX

Application

Linux

Application

Linux

Dedicated Servers

Lower utilizationHigher costMore electricity

Move workloads intoPower Systemslogical partitions

Consolidated images share resources on a single

Power System

Higher utilizationLower costLess electricity per unit of work

VIOS

Application

Linux

Application

Linux

Application

AIX

Application

AIX

Power Systems SMP Server

POWER Hypervisor


POWER7 Can Virtualize These Large Workloads, Sun Servers Can Not

45 Workloads

45 comparable Sun servers with 8 coresX4270, T5140, T5440, M5000

How many servers are needed to run these workloads?

IBM WebSphereApplication Server

2 Power 750 (3.55GHz) servers

1 64-way Power 770 (3.1GHz) server

Run 45 online banking workloads, each driving 745

transactions per second…45 LPARs

…45 LPARs

IBM Internal Study


POWER7 Costs Less To Run The Same Workload As Sun Servers (Detail)

$15329k$3984$3984k$1887k$428k$513kSoftware Purchase

and Maintenance

$1.09

$88k

$575k

Power 750(2 servers)

Sun SPARC M5000

(45 servers)

Sun SPARC T5440

(45 servers)

Sun SPARC T5140

(45 servers)

Sun Fire X4270

(45 servers)

Power 770(1 server)

$8011k$2376$1977$790k$455kHardware Purchase

and Maintenance

$23.46$6.84$6.08$2.80$1.063-Year Total Cost,

millions USD

$122k$122k$122k$122k$88kNetworking Purchase

and Maintenance

IBM Internal Study


Oracle VM For SPARC (Logical Domains) Does Not Share CPU Resources

� Threads are dedicated to each domain� Domains cannot share CPU resources with other domains automatically

4 Cores32 Threads

MortgageApplication

BankingApplication

InvestmentApplication

LoanApplication

4 Cores32 Threads

4 Cores32 Threads

4 Cores32 Threads

Domain 1

Server Utilization

Sun UltraSPARC16 Cores 8 Threads per core128 Total Threads

75%

25%

BankingApplication

Lost BusinessRequests cannot beprocessed fast enough

Domain 2 Domain 3 Domain 4

SolarisSolarisSolarisSolaris

� Utilizes 3 processor cores

� Requires 8 cores to process all requests, however domain only has 4


PowerVM Allows for Sharing of CPU Resources� Partitions can share CPU resources from a shared pool

� Improves application performance and improves server utilization

MortgageApplication

BankingApplication

InvestmentApplication

LoanApplication

Power Hypervisor

Server Utilization

75%

50%

16 Cores in Shared Pool

POWER716 Cores

AIXAIXAIXAIX

� Utilizes 3 processor cores

� When workload spikes, application uses 8 cores from shared pool


PowerVM Leads In Resource Sharing Capabilities Driving Higher Utilization And Consolidation

NoYesMemory over-commit

NoYesUncapped support

NoYesDedicated Shared mode

NoYesProcessor over-commit

Yes

Yes

Yes

Yes

PowerVM

NoMemory sharing

NoMemory compression

No

Partial

Oracle VM for SPARC (LDom)

Processor sharing (Shared Pools)

Micro-partitioning

Resource Sharing and

Resource Allocation

Factors


IBM Sub-Capacity Software Pricing Favors Virtualization On Power Systems

� How many cores must be licensed for each software product?

� On Power Systems:

� Determine the lesser of virtual processor or total physical cores in each shared processor pool

− Capped partitions may further reduce total virtual processor per pool

� Add the total pools to determine the number of cores that have to be licensed

� Oracle sub-capacity pricing NOT available for UltraSPARC T Series servers when using Logical Domains!

� Oracle does not recognize Logical Domains for sub capacity pricing

� Must license each processor core on physical server regardless of domain configuration


How Many Licenses Are Required?

� Oracle DB Licenses:� 32 VP (threads) or 4 cores on 8

Core (64 thread) systemLicense 8 cores

� Oracle WebLogic Licenses:� 32 VP (threads) or 4 cores in 8

Cores (64 thread) system License 8 cores

Sun T5440 - Sun UltraSPARC T2+

8 Cores – 64 threads

Sun Solaris + Logical Domain Manager

OracleDB

Server 1VP=16 W

LS

1-

VP

=8

WL

S 2

-V

P=

8

WL

S 3

-V

P=

8

WL

S 4

-V

P=

2

WL

S 5

-V

P=

1

WL

S 6

-V

P=

1

WL

S 7

-V

P=

1

WL

S 8

-V

P=

1

WL

S 9

-V

P=

1

WL

S 1

0 -

VP

=1Oracle

DB

Server 1VP=16

� DB2 Licenses:

�8 VP in a 4 Core PoolLicense 4 cores

� WebSphere Licenses:

�10 VPs in a 4 core PoolLicense 4 cores

8 Cores – Power 750

Shared Pool 14 Cores

Uncapped

DB2

Server 1VP=4EC=1

Uncapped

DB2

Server 2VP=4EC=1

Shared Pool 24 Cores

WA

S 1

V

P=

1

WA

S 2

VP

=1

WA

S 3

VP

=1

WA

S 4

VP

=1

WA

S 5

VP

=1

WA

S 6

VP

=1

WA

S 7

VP

=1

WA

S 8

VP

=1

WA

S 9

VP

=1

WA

S 1

0 V

P=

1


More Flexible Terms On Power Systems Saves 48% On Licensing Over Oracle/Sun

� Sun T5440 Server

� Using Logical Domains for partitioning

� Oracle DB: 8 Cores X .5 (core factor for T2+ processor)

� WebLogic: 8 Cores X .5 (core factor for T2+ processor)

� Oracle DB = $21,350

− Standard Edition

− $17,500 + $3850

� WebLogic = $12,200

− Standard Edition

− $10,000 + $2,200

� Cost:

− Oracle DB: $85,400

− WebLogic: $48,800

− Total: $134,200

� Power 750 Server

� DB2: 400 PVUs

− 4 Cores X 100 PVUs

� WAS: 400 PVUs− 4 Cores X 100 PVUs

� Cost:− DB2 WG: $50,400

− WAS SE: $19,100

− Total: $69,500

Prices as of 03/18/2010http://www.oracle.com/corporate/pricing/pricelists.html


Service Oriented Finance Needs To Meet Service Level Agreements And Provide Non-Stop Operation

I need to better meet our Service Level Agreements and have the ability to roll

out new applications without shutting my existing applications down.

CIO IBM

WebSphere can do this also! WeCall this “performance management and “non-stop operation”

Let me prove it to you!


WebSphere Virtual Enterprise Solves Three Enterprise Requirements

1. Non-stop operation� Continuous availability during application maintenance

2. Performance Management� Define service policies based on response time goals� Dynamic cluster capability allocates processor capacity to the

workloads to enforce service policies� On demand router sequences incoming requests based on response

time objectives� Ensure differentiated response time objectives for classes of customers

and applications� Maintain response time objectives despite variable workload demands

3. Efficiency� Maximize server utilization� Maximize application performance

− Throughput− Response time


quiesce & stop

Demo: WebSphere Virtual Enterprise Non-stop Operation

Edition 1.0

Edition 1.0

On-demand router

Edition 2.0

restart

application

requests

Edition 2.0

quiesce & stop

restart

Active SessionRecovery

WebSphere Application Server

Node 1

WebSphereApplication Server

Node 2

� Deploy new applications with lower risk of losing service.

� Deploy application versions without interruption.


Response Time Management Using Service Policies Assures Best Service

� Define service level goals with service policies

� Service policies specify the response time goals and the relative importance of the service policy relative to other service policies

� Application requests are mapped to service policies based on rules that classify, prioritize, and intelligently route requests to ensure that service policies are met Service Policies define the relative importance

and response time goals of application services

Oracle and Microsoft cannot do this!


WebSphere Virtual Enterprise Sequences Requests To Satisfy Differentiated Response Time Objectives

Response Time

1.5H

H

L

Response time objective is 1 second

Response time objective is 2 seconds

Both requests require 0.5 secondto process.

H

L2 L1

Response timeobjective not met

L2 L1

1.0 0.5x

Incoming Request Queue


WebSphere Virtual Enterprise Sequences Requests To Satisfy Differentiated Response Time Objectives

Response Time

1.5

L1

H

L

Response time objective is 1 second

Response time objective is 2 seconds

Both requests require 0.5 secondto process.

H

L2H

Response timeobjective not met

All response timeobjectives met

L2 L1

1.0 0.5x

Prioritized Request Queue

0.5

HL2 L1

1.5 1.0On

DemandRouter

Incoming Request Queue


Shared Resource PoolWVE On Demand Router-Classify-Prioritize-Dispatch

Stock Trade

Stock Trade.

High PriorityRequests

Low PriorityRequests

ServicePolicies

Client Requests

Stock TradeDynamic Cluster

AccountMgmt.DynamicCluster

Create Service Policies That Provides Best Service For Most Valuable Customers

1. Create Service Policies defined as response time goals2. Create rules that map application requests to service

policies based on contents of request

LP

AR

LP

AR

LostBusiness

LostBusiness

SilverServicePolicy

GoldServicePolicy

1200 ms

2000 ms

If Account Balance >= $100K thenGold Service Policy

If Account Balance < $100K then

Silver Service Policy

GOLDAverage

Response Time< 1200 ms

Service Policies

SilverAverage

Response Time

< 2000 ms


Shared Resource PoolWVE On Demand Router1. Classify Requests2. Prioritize Requests3. Dispatch Requests

LostBusiness

LostBusiness

Stock Trade

Account Mgmt.



ServicePolicies

Client Requests



WebSphere Virtual Enterprise -Maintain Response Time Objectives

ResponseTimeGoal

ResponseTimeGoal

1200 ms

StartApp2000 ms

Response TimeIncreases Due

To Demand

High priority workload maintains

response time objectives

despite increasing demand


Shared Resource PoolWVE On Demand Router1. Classify Requests2. Prioritize Requests3. Dispatch Requests

LostBusiness

LostBusiness

Stock Trade

Account Mgmt.



ServicePolicies

Client Requests



WebSphere Virtual Enterprise -Maintain Response Time Objectives

ResponseTimeGoal

ResponseTimeGoal

1200 ms

2000 ms

StartApp

High priority workload maintains

response time objectives

despite server failure


Ideal Behavior Of An Efficient System

� In an environment with multiple applications experiencing different workloads an ideal system would exhibit the following properties:

� Maximize throughput

� Minimize response time

� Capable of utilizing all available system resources with minimum overhead

� Equitable resource sharing

− Requests with same priority should have similar throughput and response time

WebSphere Virtual Enterprise can achieve these properties!

26


Comparison of Aggregate Throughput

0

50

100

150

200

250

1) WAS Pinned 2) WAS ND Pinned 3) WVE

Th

rou

gh

pu

t

Aggregate Workload

WebSphere Virtual Enterprise Throughput Is Best And Utilization Is Highest

20%

40%

60%

80%

100%

200

162157

Serv

er

Uti

liza

tio

n

63%70%

82%

and Server Utilization

Utilization

Download white paper describing study:

https://www14.software.ibm.com/webapp/iwm/web/signup.do?source=sw-app&S_PKG=wveaddsvalue&S_TACT=109HE36W&S_CMPve


WebSphere Virtual Enterprise Response Time Is Best – Up To 31% Better Than Other Scenarios

WVE has lowestaggregate

response time

Comparison of Response Times

0

0.1

0.2

0.3

1a) WAS Pinned 2a) WAS ND Pinned 3) WVE

Resp

on

e T

ime (

seco

nd

s)

Aggregate Workload


WebSphere Virtual Enterprise Delivers The Best Performance And Non-Stop Operation

Total transactionsper second

Non-Stop OperationWebSphere

Virtual Enterprise

Oracle WebLogic Server Enterprise

3375

2150

Yes

No

POWER7- 8 coresWith virtualization

HP Nehalem - 8 coresWith VMware

Three applications:

equal priority,

different demand

Workload A, 1x demand

Workload B, 3x demand

Workload C, 90x demand

Workload A, 1x demand

Workload B, 3x demand

Workload C, 90x demand


WebSphere Virtual Enterprise and Server Virtualization Together Provide Most Flexible Dynamic Infrastructure

WLM for any workload

Lower Operational Costs

Improved Application Management

Service Level Management

Application High-availability

HW high-availability

Dynamic Adjustment

Improved H/W Mgmt

Reduced Hardware �

�

�

�

�

�

�

�

�

�

�

��

�

�

�

�

�

�

�

�

�Adjust resources (i.e. CPU)Adjust placement of WAS serversRoute requests

� �


Service Oriented Finance Needs To ControlGrowing IT Labor Costs

My labor costs are going through the

roof!

How can I better control the ever

rising costs?

CIO IBM

Virtualization, standardization and automation are key to controlling rising labor costs!

WebSphere Cloudburst can solve this problem!


Leverage WebSphere Cloudburst Appliance For Faster, More Flexible Deployments

� New class of hardware appliance for data centers

� Secure, self-service cloud management device

� Dispenses hardened WAS patterns into a pool/cloud of virtualized hardware running a supported hypervisor (e.g. VMware)� Patterns leverage 10 years of WAS best

practices deployment and configuration experience

� Enables consistent and repeatable deployment of application environments based on patterns (e.g. single server, multi-tier, etc.)

WebSphere CloudburstAppliance


Deployment Study: WebSphere Cloudburst vs. Manual Installation

WebSphere CloudBurst Appliance

IBM eServer x365 Dual Core Intel Xeon3.0 GHz 12 GB Memory

VMware ESX 3.5

VM1

Gigabit Ethernet

IBM eServer x365 Dual Core Intel Xeon3.0 GHz 12 GB Memory

VMware ESX 3.5

VM1

Manual Install

33


WebSphere Cloudburst Appliance Is Fastest!

Minutes To Install Websphere Single Server

139

65

13

0

50

100

150

Manual Install Cloudburst 1st Time Cloudburst 2nd Clone

Deployment Approaches

Min

ute

s

91%less

53%less

80%less

34


Deployment Details From 1st Manual Install

10 minsDeploy Application using WebSphere Admin Console

139 MinutesTOTAL TIME TO RUN APPLICATION MANUALLY

4 minsRestart WebSphere

4 minsInstall Web 2.0 Feature Pack

3 minsDownload/Install ISA Lite

15 minsInstall WebSphere + JDK Fixpack 3

4 minsInstall SCA Feature Pack + Fixpack

1 MinInstall WebSphere Update Installer

3 MinsInstall Web Server Plugin

3 MinsInstall IBM HTTP Server

7 MinsInstall WAS 7 (including unzip)

45 MinsUpload and update Virtual Image to ESX Server

40 MinsSLES 10 SP2 Linux OS Install

TimeDeployment/Install Task

35


Deployment Details From 1st Cloudburst Install


30 SecondsSave pattern

46 minutesDeploy new Pattern with enterprise application

30 SecondsUpdate pattern with Single Server Image + Script

30 SecondsCreate new pattern

30 SecondsAdd parameters to script

1 MinuteUpload enterprise application package

1 MinuteCreate enterprise application package

15 minutesConnect Cloudburst with VMware Server


36


Deployment Details From 2nd And Subsequent Cloudburst Install


0 SecondsSave pattern

13 minutesDeploy new Pattern with enterprise application

0 SecondsUpdate pattern with Single Server Image + Script

0 SecondsCreate new pattern

0 SecondsAdd parameters to script

0 MinuteUpload enterprise application package

0 MinuteCreate enterprise application package

0 minutesConnect Cloudburst with VMware Server


37


Service Oriented Finance –Enterprise Application Overview

ClientJMS

Queue

VMware Image

WebSphere Application Server V 7

� Clients submit Loan Application form using web application� Application form stored in JMS Queue for further processing� Application running on WebSphere Application Server on VMware

38


WebSphere CloudBurst Appliance

PrivateCloud

Demo: Deploying A New WebSphere System In The Cloud

SelectPattern

CloudBurst Downloads CustomizationInformation

CloudBurst DownloadsApplication

ESX

CloudBurst Issues Deploy Image

Command

ESX Installs WAS HE Virtual Machine

Linux

WASESX AppliesCustomization

WAS InstallsApplication

WAS HE

VM Image

CloudBurst Downloads Virtual Image


# of SoftwareImagesCloneFactor

C

Virtualization, Standardization And Automation Can Reduce Labor Costs

Total SoftwareLabor Hours

Total HardwareLabor Hours

The more images you can standardize and clone, the lower you can drive software

labor hours

The greater the consolidation you can achieve, the

lower you can drive hardware labor hours

SoftwareStack

Labor HoursTotal

Labor Hours(per year)

=Hardware

Labor Hours

# of Physical Servers

+32 Hrs (Intel)64 Hrs (Power) 36 Hrs

Automation can reduce this

laborcomponent

Note: labor model best fitted to field data


0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

8,000

9,000

10,000

Total Hardware And Software Labor Hours For 45 Workloads Over 3 Years

To

tal H

ard

ware

an

d S

oft

ware

Lab

or

Ho

urs

Over

3 Y

ears

67%less

96%less

Virtualized +Standardized

(C=5) +Automation

Distributed(Intel)

Virtualized(POWER)

Virtualized +Standardized

(C=5)

45%less

77%less

One full time equivalent person (FTE) = 2080 labor hours per year

Take Your Infrastructure To The Next Level Of Agility And Cost Savings–Dynamic Application Infrastructure

Technology

energyand

service oriented

share cpu

shared resource

response time

response time

websphere

ibm internal