© Copyright IBM Corporation, 2008 Proibida cópia ou divulgação sem permissão escrita do CMG Brasil. Relative Capacity Joseph Temple – Distinguished Engineer

Proibida cópia ou divulgação sem permissão escrita do CMG Brasil. © Copyright IBM Corporation, 2008

Relative Capacity

Joseph Temple – Distinguished EngineerEduardo Oliveira – Executive IT Specialist

© Copyright IBM Corporation, 2008


TrademarksThe following are trademarks of the International Business Machines Corporation in the United States, other countries, or both.

The following are trademarks or registered trademarks of other companies.

* All other products may be trademarks or registered trademarks of their respective companies.

Notes: Performance is in Internal Throughput Rate (ITR) ratio based on measurements and projections using standard IBM benchmarks in a controlled environment. The actual throughput that any user will experience will vary depending upon considerations such as the amount of multiprogramming in the user's job stream, the I/O configuration, the storage configuration, and the workload processed. Therefore, no assurance can be given that an individual user will achieve throughput improvements equivalent to the performance ratios stated here. IBM hardware products are manufactured from new parts, or new and serviceable used parts. Regardless, our warranty terms apply.All customer examples cited or described in this presentation are presented as illustrations of the manner in which some customers have used IBM products and the results they may have achieved. Actual environmental costs and performance characteristics will vary depending on individual customer configurations and conditions.This publication was produced in the United States. IBM may not offer the products, services or features discussed in this document in other countries, and the information may be subject to change without notice. Consult your local IBM business contact for information on the product or services available in your area.All statements regarding IBM's future direction and intent are subject to change or withdrawal without notice, and represent goals and objectives only.Information about non-IBM products is obtained from the manufacturers of those products or their published announcements. IBM has not tested those products and cannot confirm the performance, compatibility, or any other claims related to non-IBM products. Questions on the capabilities of non-IBM products should be addressed to the suppliers of those products.Prices subject to change without notice. Contact your IBM representative or Business Partner for the most current pricing in your geography.

Adobe, the Adobe logo, PostScript, and the PostScript logo are either registered trademarks or trademarks of Adobe Systems Incorporated in the United States, and/or other countries.Cell Broadband Engine is a trademark of Sony Computer Entertainment, Inc. in the United States, other countries, or both and is used under license therefrom. Java and all Java-based trademarks are trademarks of Sun Microsystems, Inc. in the United States, other countries, or both. Microsoft, Windows, Windows NT, and the Windows logo are trademarks of Microsoft Corporation in the United States, other countries, or both.Intel, Intel logo, Intel Inside, Intel Inside logo, Intel Centrino, Intel Centrino logo, Celeron, Intel Xeon, Intel SpeedStep, Itanium, and Pentium are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States and other countries.UNIX is a registered trademark of The Open Group in the United States and other countries. Linux is a registered trademark of Linus Torvalds in the United States, other countries, or both. ITIL is a registered trademark, and a registered community trademark of the Office of Government Commerce, and is registered in the U.S. Patent and Trademark Office.IT Infrastructure Library is a registered trademark of the Central Computer and Telecommunications Agency, which is now part of the Office of Government Commerce.

For a complete list of IBM Trademarks, see www.ibm.com/legal/copytrade.shtml:

*, AS/400®, e business(logo)®, DBE, ESCO, eServer, FICON, IBM®, IBM (logo)®, iSeries®, MVS, OS/390®, pSeries®, RS/6000®, S/30, VM/ESA®, VSE/ESA, WebSphere®, xSeries®, z/OS®, zSeries®, z/VM®, System i, System i5, System p, System p5, System x, System z, System z9®, BladeCenter®

Not all common law marks used by IBM are listed on this page. Failure of a mark to appear does not mean that IBM does not use the mark nor does it mean that the product is not actively marketed or is not significant within its relevant market.

Those trademarks followed by ® are registered trademarks of IBM in the United States; all others are trademarks or common law marks of IBM in the United States.


Agenda

• General Ideas• Pfister’s Diagram• Workload Type/Characteristics• Workload Factor• Benchmarks• White Space• Virtualization and Utilization• Relative Capacity• Ideas International/Performance Comparison• Non-Functional requirements• Total Cost of Ownership (TCO)• Consolidation: Case Study• Platform Choices


General Ideas• It is the nature of computers that any

computer can be programmed to accomplish any task.

• Conversely, functionality is not enough.

• Rational Platform Selection is based on the solution’s “non functional requirements”


Syn

chro

niza

tion

Tim

e

Bulk Data Traffic

Shared NothingHigh LatencyBlades, Clusters,SquadronsHVRead Only Webserving, some DSS

Shared MemoryLow Medium LatencyF6800,rx8400,rp8400P670, Squadrons MLOLTP, Legacy SMP

Shared MemoryHigh Medium LatencyF12000,F15000, SuperDome,P690Data Warehouse, some DSS

From: In Search of Clusters, The ongoing battle in lowly parallel computingby Greg Pfister, p461

Shared EverythingLow LatencyzSeries, Squadrons HE OLTP, Mixed Workload

Price/

Perfo

rman

ce

Tot

al C

apac

ity

WLM

& B

I Fun

ctio

n

Virtua

lizat

ion

ArchitecturalDivide

Blades

Midrange Client Server

MainframesArchitecturalDivide

High End UNIX Severs

Simplifying the Client Server Build Out


From: In Search of Clusters, The ongoing battle in lowly parallel computing by Greg Pfister, p461


Workload Characterization

2. I/O Bound – e.g. high I/O content applications

9. Protocol Serving – e.g. static HTTP, firewall, etc.

3. Mixed Low – e.g. multiple, data-intense applications or skewed OLTP, MQ

1. Data Intensive – large working set and/or high I/O content applications

4. Mixed High – e.g. multiple, cpu-intense simple applications

8. Skewless OTLP – e.g. simple and predictable transaction processing

7. Java Heavy – e.g. cpu intensive java applications

6. Java Light – e.g. data intensive java applications

5. Database – e.g. Oracle DBMS or dynamic HTTP server

10. CPU Intensive – e.g. numerically intensive, etc.

I/O Driven

CPU Driven



z

p

Large Processors

Distributed Processing

Industry Benchmarks

TPC-C, TPCE??

Parallel Hell positioning is empirical and folklore driven


Workload ConsiderationsWorkloads may benefit from being physically close to the data

Most servers run in low to medium utilization

High I/O workloads Vs. High CPU workloads

Workloads may require high availability• Mission critical applications


Comparing servers using relative capacity:

Given system B with capacity CB processing a workload at utilization UB

capacity CA needed by system A to process the same workload is given

by:

WLF

UCC BB

A

where WLF is the Workload Factor.

With WLF we try to compensate for all the architectural differences between system A and system B. It is simplified: Actually WLF = f(UB)


Parallel

Syn

chro

niza

tion

Tim

e

Bulk Data Traffic

Shared NothingHigh LatencyBlades, Clusters,Squadrons HVRead Only Webserving, some DSS

Shared MemoryLow Medium LatencyF6800,rx8400,rp8400P670, Squadrons MLOLTP, Legacy SMP

Shared MemoryHigh Medium LatencyF12000,F15000, SuperDome,P690

Data Warehouse, some DSS


Shared EverythingLow LatencyzSeries, Squadrons HE OLTP, Mixed Workload

Pfister’s Paradigm and Core Design

Tot

al C

apac

ity

WLM

& B

I Fun

ctio

n

Serial

Processor

SpeedCache

RAS

Processor

SpeedCache

RAS

Processor

Speed

Cache

RAS


Industry Standard Benchmarks System Level Characteristics• No disk I/O• No network I/O• No database• Data sharing• Global memory• Interconnect important

Examples• SPEC OMPL2001• SPEC OMPM2001

Component

Characteristics• No disk I/O• No network I/O• No database• No data sharing• Cache / local memory• Scales w/QTY of cores

Examples• SPECint2000• SPECint_rate2000/2006• SPECfp2000/2006• SPECfp_rate2000• SPECjbb2000/2005

Specialty

Characteristics• Disk I/O• Network I/O• No data sharing• Local memory• Scales w/QTY of cores

Examples• SPECweb2005• NotesBench• SPECjAppServer 2004• SPECsfs97_R1.v3

Database Examples• TPC/H (Read Only)• SAP SD 2-Tier (Limited I/O)

System Level Database

Characteristics• Disk I/O• Network (except batch)• Global memory• Data sharing• Read/Write database

Examples• TPC/C• SAP SD 3-Tier• Oracle Apps• Oracle RAC• Oracle Batch• PeopleSoft Financials

Memory Memory I/O

CPU

Cache

CPU

Cache I/O

Local Interconnect

Memory Memory I/O

CPU

Cache

CPU

Cache I/O

Local Interconnect

Global Interconnect

Memory Memory I/O

CPU

Cache

CPU

Cache I/O

Local Interconnect

Memory Memory I/O

CPU

Cache

CPU

Cache I/O

Local Interconnect

Global Interconnect

Memory Memory I/O

CPU

Cache

CPU

Cache I/O

Local Interconnect

Memory Memory I/O

CPU

Cache

CPU

Cache I/O

Local Interconnect

Global Interconnect

Memory Memory I/O

CPU

Cache

CPU

Cache I/O

Local Interconnect

Memory Memory I/O

CPU

Cache

CPU

Cache I/O

Local Interconnect

Global Interconnect


System Design and Benchmarks

• Industry standard benchmarks are used by vendors to establish performance or price/performance leadership.

– Benchmarks are chosen to show off a platform not to allow comparisons

• Benchmarks frequently do not match real customer workloads

– Small – very limited stress on data delivery infrastructure

– Throughput oriented with highly paralyzed processing – scales with quantity of processors for all vendors

• Real workloads (including virtualization and mixed workloads) place tremendous stress on cache and system interconnect.

Wo

rklo

ad /

Ser

ver

Siz

e

Data Sharing / Workload Complexity

On-Chip CacheQty of Threads

(1-2 Sockets)Any Benchmark

Quantity ofCores / threads

Parallelization

ThroughputResults

TPC-HSAP SD 2-TierSPECJBB2005SPECintRateSPECfpRate

SPECweb

System Interconnect Cache Architecture

Schedulers # of Processors

TPC-C SAP SD 3-Tier

InterconnectCache

Schedulers# Processors

VirtualizationMixed Workload


'White space' = wasted capacity

Shared SystemsSeparate Dedicated Systems


Peak and Average

• The desired peak utilization is the “Utilization at Saturation design point”, Usd

• Average Utilization is:

sAP

Usd

sAPU

)1/(1)1/(1

1

Where:P is the Peak Load A is the Average Loads is the number of servers used to implement the capacity


Virtualization enables higher CPU Utilization

• Single workload model assumptions:– Average Utilization: 20.7%– Peak: 79%

• As more copies of this workload are added, average utilization approaches peak

– 8:1 39% Average, peak 76%– 16:1 48% Average, peak 78%– 64:1 61% Average, peak 78%

• As workload is added the number of CPUs required for the work grows at a much lower rate.

Single Appliation Server (2 CPUs)

0%

10%

20%

30%

40%

50%

60%

70%

80%

8 to 1 Consolidation (8 CPUs)

0%

10%

20%

30%

40%

50%

60%

70%

80%


0%

10%

20%

30%

40%

50%

60%

70%

80%


0%

10%

20%

30%

40%

50%

60%

70%

80%


Why Larger Servers for Virtualization?

• Hardware Advantages– Higher utilization due to shared headroom.– More internal bandwidth to improve performance– Fewer disk & network adapters and ports– Able to share memory more effectively– More fault tolerant features

• People Advantages– Fewer servers to order, install, track, maintain,

and retire– Fewer Hypervisor instances to manage– Fewer firmware patches to apply

• Data Center Advantages– Better power utilization– Reduced floor space

Hardware CapacityUsable VM Capacity

Smaller Servers

Medium Peak to Avg. Utilization Gap

Larger Servers

Small to Very Small Peak to Avg.

Utilization Gap


Relative Capacity Criteria

• What is the number and utilization of servers? – How big is this? What is the potential for virtualization or workload

management? Need to profile utilization by intervals.• How Parallel is the work?

– Read only partitioned data, mostly read minimum sharing, mostly read shared data, read/write partitioned data, read/write shared data, etc. (Leverage of shared v. partitioned resources)

• How large are the working sets for the DB, Memory, and Cache? – What is the "Workload Factor?" Need throughput v. utilization by

intervals• What are the testing and QA practices?

– How much nonproduction hardware is there as a result?


Relative capacity (cont’d)• Capacity Metric can vary (MIPS, MHz, tpm, tpc-c, n° engines, ...)

• Utilization (%) can be measured with various tools (vmstat, top, Task Manager, ...)

• WLF is measured in [CB]/[CA] units– i.e., LSPR values are MIPS/MIPS workload factors for two zSeries machines at same

utilization

• WLF difficult to measure!– not enough benchmarks to cover all the cases– driven by cache miss rate, which cannot be directly measured– “Cloud of uncertainty” around measured values


Ideas International

• Independent Organization• Publishes the Consolidated Analysis Report

(CAR)• CAR contains the RPE2 performance index• RPE2 index can be used into cross-platform

selection• Requires a License

• http://www.ideasinternational.com/ • RPE2 = Relative Performance Estimate 2 (Copyright © 2008 - Ideas

International Limited)


Performance Comparison

• Server Utilization

• Workload Type

• I/O Latency

• Cache

• Clock Speed

• Architecture

• Non-functional requirements


Cost is a quantification of Non functional requirements

• Costs go way beyond Hardware, Software, Maintenance• There are different ways to surface them• Here is one way to organize things:

– Cost of outages– Prioritization– Growth– Administration Costs– Time to Market v Code Quality– Project Costs– Environmental Costs


A full range of TCO factors considerations – often ignored

• Integration– Integrated Functionality vs. Functionality to be

implemented (possibly with 3rd party tools)– Balanced System– Integration of / into Standards

• Further Availability Aspects

– Planned outages

– Unplanned outages

– Automated Take Over

– Uninterrupted Take Over (especially for DB)

– Workload Management across physical borders

– Business continuity

– Availability effects for other applications / projects

– End User Service

– End User Productivity

– Virtualization

• Skills and Resources

– Personnel Education

– Availability of Resources

• Availability– High availability– Hours of operation

• Backup / Restore / Site Recovery– Backup– Disaster Scenario– Restore– Effort for Complete Site Recovery– SAN effort

• Infrastructure Cost– Space– Power– Network Infrastructure– Storage Infrastructure

• Additional development and implementation

– Investment for one platform – reproduction for others

• Controlling and Accounting– Analyzing the systems– Cost

• Operations Effort– Monitoring, Operating– Problem Determination– Server Management Tools– Integrated Server Management –

Enterprise Wide

• Security– Authentication / Authorization– User Administration– Data Security– Server and OS Security– RACF vs. other solutions

• Deployment and Support

– System Programming• Keeping consistent OS and SW Level• Database Effort

– Middleware• SW Maintenance• SW Distribution (across firewall)

– Application• Technology Upgrade• System Release change without interrupts

• Operating Concept– Development of an operating procedure– Feasibility of the developed procedure– Automation

• Resource Utilization and Performance– Mixed Workload / Batch– Resource Sharing

• shared nothing vs. shared everything– Parallel Sysplex vs. Other Concepts– Response Time– Performance Management– Peak handling / scalability


Overview of Techline’s case study

SAR Reports

WebserverWebserver

Webserver Server Consolidation Tool

Projected Util.

VMXT

z/VM

zLinux

Apache

Capacity Tool

= Actual Util. ?


Platform Choices

Legacy

Quality of Service

Total CostApplication Platform Support

Application Structure


Documents

© Copyright IBM Corporation, 2008 Proibida cópia ou divulgação sem permissão escrita do CMG Brasil. Relative Capacity Joseph Temple – Distinguished Engineer