Adaptive Server Farms for the Data Center

Adaptive Server Farms for the Data CenterContact: Ron Sheen Fujitsu Siemens Computers, Inc

[email protected]

Sever Blade Summit, Getting the most from your blade servers, March 22, 2005

2

The challenge – over provisioned, inflexible infrastructure

Today‘s computing infrastructure – Static and unshared islands for each service

Inefficient

Costly

Over provisioned

Hard to manage

Inflexible

Utilization of UNIX/Windows servers is low (<25% over 24 hours across all servers)Source: 2003 META Group – The Data Center of the Future

3

The challenge – Cost Reduction

Labor costs - 75% of IT budgets - inefficiently applied across multiple, separate systems – mostly sustaining costs Excessive administration, upgrades, support, service, overhead

Administration and routine IT tasks are mostly manual Limited labor resources/budget focused on manual, low value tasks

IT over invests in resources for static, standalone systems Provisioning to meet peak demand Provisioning for backup systems and applications Resources duplicated; cannot be efficiently shared

4

Addressing these challengesThe tools exist to address these challenges by:

Restructuring the data center to remove the barriers that isolate the various services into their various islands

Add tools to provision and monitor services and hardware

Add automation to automatically adapt to the changing needs of the services and users.

5

Pooling and sharing of the overall resourceRemove

Boundaries

ConsolidateStorage

Establish overall management

Assign Services

Service A Service B Service C Service D Service E

ServiceServiceEE

ServiceServiceBB

ServiceServiceAA

ServiceServiceCC

ServiceServiceDD

6

Shared Storage B1 B2 B3 Bn

.…

R/3 R/3 R/3R/3

LAN2 (Storage)

LAN1 (Public)

Shared Storage

Automation for QoS, Scaling, HA, Configuration and Provisioning

Assume a typical blade server farm - multiple blades/shared storage

Add server management, provisioning service, and control logic

Define policies and install monitoring and control agents

Server ManagementAnd Provisioning

Management Cluster

Management Cluster

AA A A

Monitoring and control agents (A) provide information on application state and QoS

Clients

7

QoS Monitoring & Management

High Water Mark

Low Water Mark

Target Metric Range

Time

QoS Metric

Measured QoS metric exceeds the specified maximum acceptable valueAllocate more satellite nodes and deploy needed application to meet QoS target

Measured QoS metric is below the specified minimal acceptable value (too many resources)

Perform orderly shutdown of some instances thus reducing cost and freeing the resources for other work.

8

Continuous Services

ServiceServiceBB

ServiceServiceAA

ServiceServiceCC

ServiceServiceDD

Adaptive infrastructure provides automated, continuous service and high availability without the costs of traditional infrastructure Application or server fails:

redirect user traffic …then

restart application or server

If no restart, but application service performance is satisfactory, there is no adjustment

If service performance is not satisfactory, then provision or reallocate another application/server

Control Nodes

Control Nodes

Server ManagementAnd Provisioning

ServiceServiceEE

9

Summary of steps To implement adaptive server farms:

Establish an infrastructure with sharable server platforms, storage and networks

Implement a deployment service This can be a traditional deployment service like e.g. Altiris or Remote

Deploy Or a virtual server deployment like VMWare

Provide a highly-available management platform Monitor and collect information from all servers Correlate data and execute reaction scripts

Include agents on each server to provide data for the management platform Direct agents for monitoring application availability and performance Interfaces to application suite tools like Oracle Grid Manager

The following slides illustrate two example deployments with SAP suite and Oracle.

10

Cost Reduction – Solution with server consolidation and automation

ServiceServiceAA

ServiceServiceCC

ServiceServiceDD

ServiceServiceBB

Service AService A Service BService B Svc CSvc C Svc DSvc DControl Systems

Unshared

Utilization 10%

Peak 50%

Different peaks

Reduced investment

Higher utilization

Automation saves labor cost

Spares

Before

After: Application Service Pools Spares with Automated Provisioning

11

Server Consolidation example - SAP

Sample configuration for a SAP test environment

Images

SAP App

SAP CI

SAP Web

WIN SQL

SA

P W

eb

SA

P W

eb

SA

P A

pp

l

SA

P C

en

tra

l

Spa

re

WIN

SQ

L

SA

P W

eb

SA

P W

eb

SA

P A

pp

l

SA

P C

en

tral

Spa

re

WIN

SQ

L

SA

P W

eb

SA

P W

eb

SA

P A

pp

l

SA

P C

entr

al

Spa

re

WIN

SQ

L

SA

P W

eb

SA

P W

eb

SA

P A

pp

l

SA

P A

pp

l

SA

P A

pp

l

SA

P A

pp

l

Sample configuration for a SAP prod. environment

12

Server Consolidation example - Oracle

Images

Oracle App S

Oracle RAC

Web Cache & J2EE

Oracle RDBMS

SA

P W

eb

SA

P W

eb

SA

P A

pp

l

Ora

cle

RA

C

Spa

re

Ora

cle

RD

BM

S

Web

Cac

hJ

2EE

SA

P W

eb

SA

P A

pp

l

SA

P C

entr

al

Spa

re

WIN

SQ

L

SA

P W

eb

SA

P W

eb

Ora

cle

AP

P S

vr

SA

P C

en

tra

l

Spa

re

WIN

SQ

L

SA

P W

eb

SA

P W

eb

SA

P A

pp

l

SA

P A

pp

l

SA

P A

pp

l

SA

P A

pp

l

Oracle front end environment

Oracle decision support environment

OLTP application environment

Ora

cle

RA

C

Ora

cle

RA

C

Ora

cle

RA

C

Ora

cle

RA

C

Ora

cle

RA

C

Ora

cle

RA

C

Ora

cle

RA

C

Ora

cle

AP

P S

vr

Ora

cle

AP

P S

vr

Ora

cle

AP

P S

vr

Ora

cle

AP

P S

vr

Ora

cle

AP

P S

vr

We

b C

ach

J2E

E

Web

Cac

hJ

2EE

We

b C

ach

J2E

E

Web

Cac

hJ2

EE

Web

Cac

hJ2

EE

Spa

re

Spa

re

Spa

re

Sp

are

13

Automated Server Farm Management

Dynamic service provisioning and workload management for blade server farms Automated, mass installation of bare-metal

blade servers Automated, mass software deployment and

software updates Priority and workload-based reusage of

resources Continuous services

Benefits Reduced administration Low-cost scalability Simplified continuous availability

Images

Linux + sendmail

W2K + IIS

Linux + Apache

W2K + Citrix + MS Office

Lin

ux+

Ap

ach

e

Lin

ux+

Ap

ach

e

Lin

ux

+S

en

dM

ail

Lin

ux+

sen

dM

ail

Lin

ux+

Ap

ach

e

Win

+II

S

Win

+II

S

Spa

re

Win

+C

itrix

Win

+C

itrix

tech-1

14

Conclusion

Many of the costly operations in the data center can be automated

These techniques will provide high available and high quality of service without the overhead and complexity of traditional clustering

The following demo will illustrate how this all works together.