The Problems in Software Estimating
Post on 02-Jan-2016
52 Views
Preview:
DESCRIPTION
Transcript
BE AMERICA’S BEST
Software Technology Support Center
The Problems in Software
Estimating
Dr. Randall JensenSoftware Technology Support Center
Hill AFB, UT
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
Outline
Evolution of software estimating models
Software problems
Management is the issue
Data collection concerns
New models?
It’s all about models – “Existence is elsewhere”
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
Project Uncertainty Principle
O G D E N A I R L O G I S T I C S C E N T E R
Software model genealogy
NordonIBM 1970
Putnam/SLIMUS Army/GE/QSM 1976
DotyRADC 1977
JS1,2,3CEI 1980
SEER-SEMGAI 1989
Seer II / SageSEI 1995
COCOMOTRW 1981
COCOMO IIUSC 1995
Validation
SeerHAC 1979
PRICE-SPrice 1977
TRUE-SPrice 2003
REVICUSAF 1991
Management
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
Sage/SEER-SEM - Normal form
where K = life cycle cost (person years), Td = development time (years)
slocdtee TKCS
3dT
KD
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
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
Software problems are not new
Unreliable
Late delivery
Modification costs prohibitive
Impossible to maintain
Inadequate performance
Product exceeds budget costs
1968 CS Conference, Munich, Germany
KEY
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
Historic note:
“More software projects have gone awry for lack of calendar time than for all other causes combined…”
F. P. Brooks, Jr., Mythical Man Month
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
Common technology claim
If you use (new technology),Productivity will improve by an order of magnitude
AndDefects will reduce to zero
Note: None of the history-based estimating models (tools) support this claim, not even close.
Neither does history.
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
There is always HOPE
1960 1970 1980 1990
100
90
80
70
60
50
Ada
OOD
Structured Analysis
Structured DesignProcess Maturity
PWB
Structured Programming
3rd Generation Languages
PRODUCTIVITY
lppm
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
3 Dimensions of development
PROJECT
PROCESSPEOPLE
1975
1985
1995
1938
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
Impact of management on cost
“Poor management can increase software costs more rapidly than any other factor…
Despite this variation, COCOMO does not include a factor for management quality, but instead provides estimates which assume that the project will be well managed [italics mine, RWJ]”
Boehm, B. W., Software Engineering Economics, (Englewood Cliffs, NJ: Prentice-Hall, Inc.), 1981, pg. 486
Note: On a scale of 0 to 10 well managed must rank approximately 3.4?
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
Where would you focus effort?
3
11
17
64
TOOLS
PEOPLE
SYSTEMS
MANAGEMENT
Source: G. Weinberg, Quality Software Management, Vol. 3
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
Biased research?
Tools People Systems Managers0
10
20
30
40
50
60
70
SEI Papers
Pe
rce
nt
Source: G. Weinberg, Quality Software Management, Vol. 3
O G D E N A I R L O G I S T I C S C E N T E R
Cost driver relative impact
Target ExperienceMultiple Sites
Develop Sys ExperienceMultiple Organizations
Resource LocationDevelopment Tools
Modern PracticesApplication Experience
Practices ExperienceStaff Capability
0 0.5 1 1.5 2 2.5
Relative Development Cost Impact
It takes people to implement processes and use technology.
April 21st, 2009
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
BIG productivity drivers
Analyst capability Management style Motivation Use of team methods Working environment Problem solving skills
Programmer capabilityApplication domain experience Use of modern practicesAutomated tool support
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
Capability shift (environment)
Basic Technology Constant
1980 2005
5500 750065006200 85008000
8635(1990)
O G D E N A I R L O G I S T I C S C E N T E R
Historic data observations
Models are made from historic data More data is better Adequate historic data takes time to collect (Lots of time)
Repository vs Database Repository = a place where things may be stored for safekeeping
Data stored without concern for a formal definition of the data placed in the repository
SRDR is defined as a financial report that happens to contain technical information
Database = a collection of data arranged for ease and speed of retrieval
Requires consistency of entries Each data field must satisfy common definition
DCARC is a repository as was SMC database Data definitions?
BE AMERICA’S BEST 18
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
Problems?
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
Popular estimating solution
Let’s create a new software model
Models are out of date
Models are inaccurate
Too difficult to use
Oh, really! Are you sure?
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
New estimating models?
Lack of confidence in existing tools (current issues) Optimistic estimates Unacceptable estimates
Aging tools (?) Culture is constant New models require validation
New models -- No silver bullets Quality data (if, where, proprietary) New models require validation
New (or better) estimators? Experience Skill Integrity
O G D E N A I R L O G I S T I C S C E N T E R
EstimatingPrediction is very hard, especially when it’s about the
future
top related