34
Fit For Purpose Joe Temple IBM [email protected] 8/9/2011 Session 09772 System z System x Power Time Horizon ISV Support Non-functional Requirements Environmental Constraints Strategic Direction TCO Model Skills Politics Architecture Technology Adoption Deployment Model Scale Geographic Considerations

Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

  • Upload
    others

  • View
    4

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Fit For PurposeJoe TempleIBM [email protected]

8/9/2011Session 09772

Systemz

System x

Power

Time HorizonISV Support

Non-functionalRequirements

EnvironmentalConstraints

Strategic Direction

TCO ModelSkills

Politics

Architecture

TechnologyAdoption

DeploymentModel

Scale

GeographicConsiderations

Page 2: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Key IBM Initiatives• Smarter Planet• Smart Computing• Big Data• Workload Optimization• Cloud• Fit for Purpose

Page 3: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

What is F4P?Fit for Purpose is a client centric thought process that when applied yields rational

platform choices which are in line with the client’s requirements and local conditions.

It is based on the fundamental principles that “one size does not fit all” and that “local factors matter”.

Systemz

System x

Power

Time HorizonISV Support

Non-functionalRequirements

EnvironmentalConstraints

Strategic Direction

TCO ModelSkills

Politics

Architecture

TechnologyAdoption

DeploymentModel

Scale

GeographicConsiderations

Page 4: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Systemz

System x

Power

Time HorizonISV Support

Non-functionalRequirements

EnvironmentalConstraints

Strategic Direction

TCO ModelSkills

Politics

Architecture

TechnologyAdoption

DeploymentModel

Scale

GeographicConsiderations

Plat

form

cap

abilit

y

Platform A Platform B

Local Factors

Local Factors

Local factors affect platform selection

Apollo 13

Local Factors Matter• Skills• Technology adoption• Management• Volume of servers• Organizational

Scale Matters• Changes people dynamics• Increases handoffs• Affects testing, patching, etc

Business Applications

Transaction Processing and Database

Web, Collaboration and Infrastructure

Analytics and High Performance

Workloads Matter

In a Nutshell 1A lot to consider

Page 5: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Larger servers offer virtualization advantagesNon-functional requirements are the significant

element of platform selection

Beware of the hidden cost of sub optimizationChargeback models often distort the selection process

Cost per Transaction

Workload Volume

Cos

ts p

er T

rans

actio

n

Dedicated

Shared

TCA TCO

In a Nutshell 2

There are server design trade offs There are operational trade offs

Service Level

Throughput Efficiency

Serial Fitness

Parallel Fitness Data Fitness

All Tactical Choices

All Strategic Choices

Tactical Strategic

Complete Business

Case

IT Decision

Strategic and tactical balance matters

Page 6: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Workload Optimization:F4P is not the same thing

Page 7: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Workload Optimization – One size does not fit all

IBM has used IDC market segments to differentiate workloads

Analytics and High Performance

Transaction Processingand Database

Web, Collaboration and Infrastructure

Business Applications

Page 8: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Fit for purpose views the same segments differentlyAn SAP solution will exhibit any or all of these types

Mixed Workload – Type 1• Scales up• Updates to shared data

and work queues• Complex virtualization• Business Intelligence with

heavy data sharing

Parallel Data Structures – Type 3

Small Discrete – Type 4

Application Function Data Structure Usage Pattern SLA Integration Scale

Highly Threaded – Type 2

• Scales well on clusters• XML parsing• Buisness intelligence with

little data sharing• HPC applications

• Scales well on large SMP• Web application servers• Single instance of an

ERP system• Some partitioned

databases

• Limited scaling needs• HTTP servers• File and print• FTP servers• Small end user apps

Black are design factors Blue are local factors

Page 9: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Workload Types and Pfister’s Paradigm

How

much sharing is there?

Synchronization TrafficC

ontention and Coherence D

elays

How much data do we we have to deal with?Bulk Data Traffic – Saturation Delay

Type 1Shared data and work queues

Type 3Parallel data structures

Type 4Small discreet applications

Type 2 Highly threaded

applications

Parallel PurgatoryParallel Nirvana

Parallel Hell

Page 10: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Pfisters Paradigm: One size does not fit all

Sing

le T

hrea

ded

Wor

k

Data Load

Con

tent

ion

and

Coh

eren

ce

Saturation

ParallelNirvana

ParallelHell

ParallelPurgatory

Que

ry L

oads

Legacy SMP

Loads

Scaled up OLTP and “Mixed”

Loads

Single Threaded

Loads

with clustering

with

dat

a up

date

s

Withcode Paralleliza

tion

with Virtu

alizatio

n

with structure

Page 11: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

You can put any machine on this chartAll you need to know is:

Clock SpeedTop Level Cache Size and Sharing ScopeCore Count and Thread Count per Core

Throughput Claim for Multithreaded Cores

Workload Optimization 8 Socket MachinesBubble Size is Parallel Fitness - Thread Count

0

0.5

1

1.5

0 0.5 1 1.5

Data Fitness - Cache/Thread

Seria

l Fitn

ess

- Clk

* SM

T Ta

ke D

own

z196p780 STp780 SMT2p780 SMT4p780T STp780T SMT2p780T SMT4X7560 STX7560 HTX7542

The SMT Take down is 1/SMT throughput multiplier. This is an application of “Little’s Law”

The trend line comes out of the page on the lower left.

There is a clear trade off at work here:

To have more concurrent threads you must give up thread speed and cache/thread

Note: Each brand will define capacity in an advantageous way which is why fundamentals work better than benchmark derived metrics.

Page 12: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

ingle

IBM Systems are fit for purpose and cover key legaciesFi

tnes

s fo

r sin

gle

thre

aded

wor

k

Fitness for Data Centric Work

Enterprise SMP

z/OS®

Linux®

AIX®

Linux

i5/OS™

LinuxAIX

WindowsAIXLinuxSolaris™

HE System x™

WindowsLinuxSolaris™

IBM BladeCenter™

System p755 and p775™

Con

tent

ion

and

Coh

eren

ce

Saturation

Fitness fo

r Virtu

alizatio

n

New Nehalem Scalable Blades

WindowsLinuxSolaris™

WindowsAIXLinuxSolaris™

Choose based on Fit

zEnterprise with zBX

Page 13: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

“Hybrid Computing” turns “Client Server” on its ear

Presentation

Bus Logic

Data and State

Client Server Tiers Centralized Sharing

User NetworkUser Network

Hybrid Computing

User Network

Services

Private Connection

User Network

Page 14: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Host

Hybrid Computing

User Network

Services

Private Connection

User Network

Open CL Language– Math accelerator API

WebSphere Compute Grid – JAVA API

CoZ Launcher – Batch to secure shell API

Hybrid APIs available today

Page 15: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Cloud is not about deployment model or server type.

• Cloud is simply a layer of software• Fills gaps from underlying platform in workload management,

provisioning, etc.• Provides for self service by clients• Provides a chargeback mechanism by which clients rent services

• IAAS Infrastructure as a Service• SAA Software as a Service• AAAS Application as a Service• DAAS Data as a Service• BPAAS Business Process as a Service

Page 16: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Clouds are platforms

• When designing them there are fit for purpose decisions in choosing the underlying infrastructure

• When employing them there are fit for purpose decisions about using them

Page 17: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Data

SerialParallel

z196

p780 SMT2

p780 SMT4

p780T ST

Another way to look at the data

This leads to the notion of a “trade off triangle” Serial Fitness

Parallel Fitness Data Fitness

Page 18: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Sign in my office:

It’s about the Client, stupid

Local factors not only matter, they rule the day

Page 19: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Server positioning is not enough

• The positioning above is about the application and data design v machine design.

• What about the local factors?• Usage Pattern• SLA• Integration• Scale

Page 20: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

The Operational Tradeoff Triangle

Service Level

Throughput Efficiency

Service Level

Throughput Efficiency

Service Level

Efficiency

Throughput

Service Level

EfficiencyThroughput

The Local Factors are related to this triangle through “Normalized Headroom”

Page 21: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

The Operational TradeoffGoverned by “Normalized Headroom”

• HR = (1-u)/u = c2Nt0/twait

• HR(avg) = kcN2

• U = 1/(1+HR)

• twait/t0 = c2N/HR = c2Nu(1-u)

• twait = (t0)(c2N)(u/(1-u)) = (capacity)(variability)(utilization)M/G/1 system

Page 22: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

So how do we relate the two triangles?

Service Level

Throughput Efficiency

SerialFitness

ParallelFitness

DataFitness

Parallel Fitness primarily drives throughput , but can drive Service Level (Wait Time)

Serial Fitness primarily drives Service Level (Service Time) but will also drive throughput

Data Fitness primarily drives Efficiency by forestalling saturation, but can drive Service Level by maintaining low service time at higher loads.

Page 23: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

So there is a primary corner to corner relationship

Service Level

Throughput Efficiency

SerialFitness

ParallelFitness

DataFitness

Page 24: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

The local factors distort the Client’s Tradeoff

Service Level

Efficiency

Throughput

Serial Fitness

Data Fitness

Parallel Fitness

Page 25: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Sometimes the local factors are at odds with the design.

Service Level

Efficiency

Throughput

SerialFitness

ParallelFitness

DataFitness

This leads to over provisioning and low utilizationor under provisioning and low service level.

Page 26: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

And that is just the workload optimization story

Systemz

System x

Power

Time HorizonISV Support

Non-functionalRequirements

EnvironmentalConstraints

Strategic Direction

TCO ModelSkills

Politics

Architecture

TechnologyAdoption

DeploymentModel

Scale

GeographicConsiderations

Client support is delivered through F4P Discussions, Briefings, Workshop Offerings, and ongoing System Architect support.

Always check to see if an SA is engaged or can be engaged for this work

Page 27: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Workload is usually just one row of the F4P workshop output

2113Discipline of Skills1

6939Skills3

1311Legacy/ Momentum1

4246Security2

1123Chargeback Usage Reporting1

1123SLA enforcement/Prioritization1

4244Hi Availability2

1123Disaster Recovery1

6396Workload3

DCBAWeight

Weighted Scores

Page 28: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Which can yield value/requirement matching input and “foil” for TCO analysis

Weighted Scores

0

5

10

15

20

25

30

35

40

45

A B C D

Discipline of Skills

Skills

Legacy/ Momentum

Security

Chargeback UsageReportingSLAenforcement/PrioritizationHi Availability

Disaster Recovery

Workload Fit

Disaster Recovery

Hi Availability

SLA enforcement/Prioritization

Chargeback Usage ReportingSecurity

Legacy/ Momentum

Skills

Discipline of Skills

A B C D Template

A provides best total valueB has best value match

C is most inline with local strategy

Page 29: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Lab Services Fit for Purpose workshop work flow

Review decision criteria, platform definitions and

pre-workshop preparation

Review decision criteria, platform definitions and

pre-workshop preparation

Review Fit 4 Purpose concepts and theoryReview Fit 4 Purpose concepts and theory

Conduct on-site interviews with key

management, operation’s and sytems’ admin.

representatives of each in-scope

technology platform

Conduct on-site interviews with key

management, operation’s and sytems’ admin.

representatives of each in-scope

technology platform

Evaluate customer’s IT environment against each criteria, build

weighting structure and

complete model

Provide “turn-around’documents for customer

input to the model

Provide “turn-around’documents for customer

input to the model

Final PresentationPresent findings and

conclusions.Present platform decision

model and train on its use, maintenance and

modification

Overview of objectives, scope and final

deliverable

Overview of objectives, scope and final

deliverable

Kick Off Phase-off site

Interview Phase-on site

Work product development and final presentation

Page 30: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

The lab services model: More comprehensive, precise and broader scope

Page 31: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Lessons Learned

• The scope & impact will be strategic in nature.

• Relationships are key and everyone has to “behave”• Failed workshops have usually been caused by excessive product advocacy

• Customers like what they are seeing and are telling us they are working to determine how they will incorporate it into their own decision making process going forward.

• F4P proved to be a significant competitive advantage.

• For clients seeking answer to the question, “where do we run this application?”• Fit for Purpose is framework to help you decide. • An in depth dialog with the customer – pros and cons, tradeoffs, etc.

• Not a sales call… consultative• Identify value differentiators and weightings • Ongoing reuse of “local factors” climbs a learning curve

• Indeed one size does not fit all; we have both mainframes and “Watson” for good reasons.

Page 32: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Key Challenges

• Being Client Centric in a brand dominated environment• Requires a strong minded client first approach• Avoid advocacy: F4P is more about “listening” than “telling”

• There is a fine line between articulating value and platform advocacy• The client defines the line by the questions asked and assertions made

• Avoiding our own brand centric biases• We all come with baggage• Many things that “everybody knows” are not true

• Most of what you know about relative capacity falls into this • Avoid accepting the common wisdom about “the once and future

platform”• Need to avoid gaming the system unintentionally• Can’t tolerate those who game the system intentionally• AHP method can help

• Pairwise comparison of requirements and platforms • Can’t see overall picture until the end• Hard to apply spin until done.

Page 33: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Questions

Page 34: Fit For Purpose · 2011-08-09 · pre-workshop preparation Review Fit 4 Purpose concepts and theory Review Fit 4 Purpose concepts and theory Conduct on-site interviews with key management,

Some “Light Reading” on the Topic

• “Server Platform Selection and Positioning”, Lebsack, Dixon CMG Conference Proceedings, December, 2009..

• Dr. Gregory Pfister , In Search of Clusters, the ongoing battle in lowly parallel computing, Second Edition, Prentice Hall, 1998

• Roger Rogers and Joe Temple, “Relative Capacity and Fit for Purpose Platform Selection”, CMG Journal of Computer Management, no 123, March 2009

• Dr. Neil Gunther: Guerilla Capacity Planning

• Rick Lebsack and Joe Temple “Fit for Purpose Platform Selection, a Workload View” on IBM Techdocs and pending CMG Journal

• See Joe Temple for drafts: “The Operational Trade off Triangle”, “The Server Design Trade off Triangle” and “Using Normalized Head Room for Infrastructure Analysis and Design”