Software Technology Support Centergy pp

Why SoftwareWhy Software Projects Fail

Dr. Randall JensenLes Dupaix

Software Technology Support CenterHill AFB, UTHill AFB, UT

BE AMERICA’S BEST

Project success rateO G D E N A I R L O G I S T I C S C E N T E R

j

CHAOS 2004 S ft P j t SCHAOS 2004 Software Project Survey

Failed18%Success 18%Success

29%

Challenged53%53%

BE AMERICA’S BEST

CHAOS Success rate definitionsO G D E N A I R L O G I S T I C S C E N T E R

P d d t bl ltSuccess Produced acceptable results delivered close to on-time and on-budget

Challenged Delivered software product g psignificantly over budget and/or schedule

Failed Failed to deliver any usable result within budget orresult within budget or schedule constraints

BE AMERICA’S BEST

Software problems are not newO G D E N A I R L O G I S T I C S C E N T E R

p

UnreliableUnreliable

Late delivery

Modification costs prohibitive

Impossible to maintainp

Inadequate performance

P d t d b d t tProduct exceeds budget costs

1968CS Conference, Munich, Germany

BE AMERICA’S BEST

Reasons for project failureO G D E N A I R L O G I S T I C S C E N T E R

p j

ADVERTISEDADVERTISEDImmature technology

Inadequate requirements

Insufficient developer experienceInsufficient developer experience

COVERTPoor estimating and planning

Hope (Pandora’s paradox)Hope (Pandora s paradox)

BE AMERICA’S BEST

Repeatable thingsO G D E N A I R L O G I S T I C S C E N T E R

p g

Development environmentDevelopment environmentProductivityMi i d l t tiMinimum development time

Effective product sizeC l itComplexityPaul Masson rule

Let’s look at some proof

BE AMERICA’S BEST

BIG productivity driversO G D E N A I R L O G I S T I C S C E N T E R

BIG productivity drivers

A l t bilitAnalyst capabilityManagement styleM ti tiMotivationProblem solving skillsUse of team methodsUse of team methodsWorking environment

Application domain experienceApplication domain experience Automated tool support Programmer capabilityProgrammer capabilityUse of modern practices

BE AMERICA’S BEST

Capability shift (environment)O G D E N A I R L O G I S T I C S C E N T E R

1980 2005

8635(1990)

5500 750065006200Basic Technology Constant

85008000

BE AMERICA’S BEST

Basic Technology Constant

1996/2003 Productivity for Avionics and Unmanned Space

O G D E N A I R L O G I S T I C S C E N T E R

and Unmanned Space

1.0

0.8

velo

pmen

t...

0.6

softw

are

dev

0 5

0.2

0.4

bilit

y th

at th

e

Productivity (2003)Productivity (1996)50th percentile (2003)

0.5

0.00 50 100 150 200 250 300

Prob

ab 50th percentile (2003)50th percentile (1996)

47 64

..progressed at LESS THAN this rate in lines/developer-months1996 graph based on 52 Mil-Spec Avionics & Unmanned Space Softw are Data Records2003 graph based on 51 Mil-Spec Avionics & Unmanned Space Softw are Data Records

BE AMERICA’S BEST

Aerospace Corp – Long etal 2004

1996/2003 Productivity Distributions for Ground and Mobile

O G D E N A I R L O G I S T I C S C E N T E R

for Ground and Mobile

1.0..

0.8

velo

pmen

t.

0.6

oftw

are

dev

0 5

0.4

that

the

so

Productivity (2003)P d ti it (1996)

0.5

0.2

Prob

abili

ty Productivity (1996)

50th percentile (2003)50th percentile (1996)

0.00 100 200 300 400 500 600

...progressed at LESS THAN this rate in lines/developer-months

P

105 138

1996 graph based on 112 Military Ground & Military Mobile Softw are Data Records2003 graph based on 135 Military Ground & Military Mobile Softw are Data Records

BE AMERICA’S BEST

2003 graph based on 135 Military Ground & Military Mobile Softw are Data Records

Aerospace, Long et al, 2004

Common technology claimO G D E N A I R L O G I S T I C S C E N T E R

gy

If you use (new technology),

Productivity will improve by an order of magnitude

And

Defects will reduce to zero

BE AMERICA’S BEST

There is always HOPEO G D E N A I R L O G I S T I C S C E N T E R

y

100

90Ada

Structured AnalysisPRO 90

80 OOD

Structured DesignProcess Maturity

ODUC

70

OOD

PWB

TIV

60

50

W

Structured Programming3rd Generation Languages

ITY

1960 1970 1980 199050

lppm

BE AMERICA’S BEST

Where would you focus effort?O G D E N A I R L O G I S T I C S C E N T E R

3 TOOLS

11 PEOPLE

17 SYSTEMS17

64

SYSTEMS

MANAGEMENT64 MANAGEMENT

BE AMERICA’S BEST

Source: G. Weinberg, Quality Software Management, Vol. 3

Relative payoffO G D E N A I R L O G I S T I C S C E N T E R

70

60

70

40

50 SEI Papers

cent

30

40

Perc

10

20

Tools People Systems Managers0

BE AMERICA’S BEST

Source: G. Weinberg, Quality Software Management, Vol. 3

Repeatable thingsO G D E N A I R L O G I S T I C S C E N T E R

p g

Development environmentProductivityMinimum development time

Effective product sizeComplexityPaul Masson rule

Let’s look at more proof

BE AMERICA’S BEST

Historic project dataO G D E N A I R L O G I S T I C S C E N T E R

p j

Source: Long, L. G. et al, 2004

BE AMERICA’S BEST

g

Historic space project limitsO G D E N A I R L O G I S T I C S C E N T E R

Historic space project limits

200

Successful Completions

0

Time5 years0

0

BE AMERICA’S BEST

Source: Long, L. G. et al, 2004

Three development variablesO G D E N A I R L O G I S T I C S C E N T E R

p

Cost

Schedule

Scope

Developer can control any two

BE AMERICA’S BEST

p y

Historic note:O G D E N A I R L O G I S T I C S C E N T E R

“More software projects have More software projects have gone awry for lack of calendar time than for all other causes time than for all other causes combined…”

F. P. Brooks, Jr., Mythical Man Month

BE AMERICA’S BEST

Common estimate dilemma O G D E N A I R L O G I S T I C S C E N T E R

BE AMERICA’S BEST

Constraint analysisO G D E N A I R L O G I S T I C S C E N T E R

y

Success and failureSuccess and failure determined by

E t d t d h d lExpected cost and scheduledetermined by

Project plandetermined byy

Cost and schedule ESTIMATESdetermined bydetermined by

Managers and/or Estimators

BE AMERICA’S BEST

Key estimate factorsO G D E N A I R L O G I S T I C S C E N T E R

y

X

Start Here

BE AMERICA’S BEST

Elements for successful estimates

O G D E N A I R L O G I S T I C S C E N T E R

estimates

Basic understanding of the requirements

Ability to accurately size the deliverable product

Assessment of the deliverable complexityAssessment of the deliverable complexity

Profile of the organization’s delivery capability

BE AMERICA’S BEST

Estimates are importantO G D E N A I R L O G I S T I C S C E N T E R

p

Without well thought out estimates, there is no firm basis for:

Defining costs and schedulesMaking tradeoffsMaking tradeoffsAssociating development costs with the benefitsConforming to a budget or scheduleConforming to a budget or schedule

THE PROJECT IS ALREADY OUT OF CONTROL!THE PROJECT IS ALREADY OUT OF CONTROL!

BE AMERICA’S BEST

Project success rateO G D E N A I R L O G I S T I C S C E N T E R

j

CHAOS 2004 S ft P j t SCHAOS 2004 Software Project Survey

Failed18%Success 18%Success

29%

Challenged53%53%

BE AMERICA’S BEST

Success definitions: 2nd lookO G D E N A I R L O G I S T I C S C E N T E R

Success definitions: 2 look

Success Produced acceptable results delivered close to on-time

Challenged

and on-budget

Delivered software productChallenged Delivered software product significantly over budget and/or schedule

Failed Failed to deliver any usable result within budget or

h d l ischedule constraints

BE AMERICA’S BEST

ConclusionsO G D E N A I R L O G I S T I C S C E N T E R

Software project failures are primarily due to failedSoftware project failures are primarily due to failed expectationsTechnology, requirements and experience are largely accounted for in resource estimatesFailure ranking:

1. Poor estimating and planning2 Hope (Not a management strategy)2. Hope (Not a management strategy)3. Immature technology 4. Inadequate requirements5. Insufficient developer capability (Ebonians)

BE AMERICA’S BEST

New estimating models?O G D E N A I R L O G I S T I C S C E N T E R

g

Lack of confidence in existing toolsLack of confidence in existing toolsOptimistic estimatesUnacceptable estimatesUnacceptable estimates

•Skill •Experience I t it• Integrity

Aging toolsQuality dataQuality data Culture is constantNew models require validationNew models require validation

Silver bulletNew estimators?

BE AMERICA’S BEST

New estimators?

Historic note:O G D E N A I R L O G I S T I C S C E N T E R

We learn from We learn from experience that we don’t learn we don t learn from experience.

D. H. Lawrence

BE AMERICA’S BEST

O G D E N A I R L O G I S T I C S C E N T E R

The End, or

is itis itThe Beginning?

(History does repeat itself)( y p )

BE AMERICA’S BEST