Upload
ngonhan
View
219
Download
0
Embed Size (px)
Citation preview
A University of Southern California
c I S I E I Center for Software Engineering - COCOMO I1 Status and Plans
Brad Clark, Barry Boehm
USC-CSE Annual Research Review
March 10, 1997
University of Southern California c S ( E ( Center for Software Engineering 81,
Outline
COCOMO I1 Status: Brad Clark - Model calibration
- Tool Status
- Data Status
COCOMO I1 Plans: Barry Boehm I
- Tool Plans
- Model Plans
- Training and data collection plans
UNIVERSITY OF SOUTHERN CALIFORNIA
Presentation Outline: *Model Calibration
Calibration Procedures
COCOMO 11.1997 Model paramiters
Accuracy Results
USC COCOMO Software Status
COSTAR Software Status
Calibration Data Status I
BKC 2./23/97 CENTER FOR SOFTWARE ENGINEERING
UNrVERSITY OF SOUTHERN CALIFORNIA
Model Calibration Status: Three models comprise COCOMO 11:
+ Applications Composition I
+ Early Design
+ Post-Architecture
Post-Architecture model calibrated
Early Design model will be derived from Post-Architecture
Applications Composition: need data I
BKC 2/23/97 CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
Calibration Process: Begin with expert-determined apriori model parameters
Collect Data
Identify and consolidated highly correlated model parameters
Statistically determine estimates of consolidated model parameters from data
Use data determined coefficients to adjust apriori model parameters
Experiment with weighting I factors
BKC 2/23/97 CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
Post-Architecture Model: Non-linear model:
1 1
PM = size)^ O ~ E M , estimated
A: Multiplicative calibration variable
B : Captures relative diseconomies of scale. Consists of 5 scale factors:
B=1.01+ S F .
EM: Effort Multipliers to reflect characteristics of particular software under development.
0 Size: Derived from either Source Lines of Code or Function Points. Includes reuse and breakage effects.
BKC 2/23/97 CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
Apriori Model Parameters:
I Driver / Symbol I VL I L I N I H I
FLEX SF2 0.05 0.04 0.03 0.02 I RESL SF3 0.05 0.04 0.03 0.02 I TEAM I SF4 1 0.05 10.04 1 0.03 10.02 1 I PMAT I SF5 1 0.05 10.04 / 0.03 )0.02 ( 1 RELY I EM, 10.75 10.88 11.00 11.15 1 I DATA 1 EM2 I I CPLX I EM, 10.7510.88/ 1.001 1.151 IRUSEI EM, I ~ ~ 0 . 8 9 ~ 1 . 0 0 ~ 1 . 1 6 ~ I DOCU I EM, ( 0.85 1 0.93 / 1 .OO 1 1 .08 1 I TIME I EM, I I
( PVOL I EM, I BKC 2/23/97 CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNU
I Driver I Symbol I VL I L N H I V H I X H I
BKC 123197
ACAP 1 EM9 1 1.5 1 1.22 1 1 .OO 1 0.83 / 0.67 1 I PCAP I EMlo ~ 1 . 3 7 ~ 1 . 1 6 ~ 1 . 0 0 ~ 0 . 8 7 / 0 . 7 4 ~ I I PCON I EM,, 11.261 1.11 11.0010.91 10.831 I
I SITE I EM,, 11.24 11.10 11.00 10.92 / 0.85 10.79 1
AEXP PEXP LTEX
CENTER FOR SOFTWARE ENGINEERING
EMlP EM,, EMl4
SCED
1.23 1.26 1.24
EM,,
1 . 1 1.12 1 1
1.23
1.00 1.00 1.00
1.08
0.88 0.88 0.9
1.00
0.80 0.80 0.82
1.04 1.10
UNIVERSITY OF SOUTHERN CALIFORNIA
Data Collection: Define the data needed (to completely describe the Post Architecture Model) .
Collect data with a paper form or a computer software tool
Affiliate Organizations providing majority of data.
+ Historical - whole project
Site visits or phone interviews to record data
Enter in data into the re~ositorv
Data is labeled with generic id
Stored in locked room
Limited access by researchers
Data Consistency checking and conditioning BKC 2/23/97 CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
Consolidated Highly Correlated Parameters :
1 TlME 1 STOR IACAP I PCAP TIME 11.0000 1
ACAP 1-0.2855 1-0.0769 I 1.0000 I
Combined (for calibration purposes only):
+ TIME & STOR into RCON (Resource Constraints)
PCAP
+ ACAP & PCAP into PERS (Personnel Factors)
-0.0027 -0.20 15
Thus, calibrated 15 effort multipliers instead of 17
BKC 2/23/97
0.7339
CENTER FOR SOFTWARE ENGINEERING
1.0000
UNIVERSITY OF SOUTHERN CALIFORNIA
Expanded Post-Architecture Model: Distribute the Scale Factors
2 1 predictor variables: 15 Effort Multiplier Coefficients + 5 Scale Factor Coefficients + overall A constant:
,S t = A ( S i z e ) E M ,-• E M 1 5
1.01
Log Transformed Model: Regression analysis will derive the coefficients, A and bi, for each factor
1
In(PM est )- ln(Si~e)~l .Ol= A+b,SF,ln(Size)+~~~+b,,ln(EMl,)
BKC U23197 CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
Example of Applying Coefficients to Model Apriori Parameters:
I
Driver I Symbol PREC I SF, FLEX I SF, RESL 1 SF, TEAM I SF, PMATI SF, RELY I EM, DATA I EM, CPLX I EM,
DOCUI EM,
PVOL I EM, PERS I EM,,, PCON 1 EM,, AEXP I EM,,
CENTER FOR SOFTWARE ENGINEERING BKC 2/23/97
UNIVERSITY OF SOUTHERNCALIFORNIA
RUSE Effort Multiplier: Example of the effect of a negative coefficient
BKC 2/23/97 CENTER FOR SOFTWARE ENGINEERING
+ 10% Regression -a- Regression Results
UNIVERSITY OF SOUTHERN CALIFORNIA
Distribution of RUSE:
Frequency 30
BKC 2/23/97
0.9 1 .O 1 .I 1.2 I .3 I .4 1.5 1.6
RUSE
CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
Evolving Model Values:
100% Data Driven
BKC 2/23/97
100% Expert Driven
Number of projects used in calibration
CENTER FOR SOFTWARE ENGINEERING
UNNERSITY OF SOUTHERN CALIFORNIA
Aposteriori Model Parameters Using 10% of data-determined and 90% of apriori
Effort constant, A: 2.45
FLEX I SF, 10.0607 10.0486 IRESL 1 SF, 10.0422 10.0338 TEAM 1 SF, 10.0494 (0.0395 IPMAT I SF, 10.0454 10.0364 RELY I EM, 10.75 10.88
~CPLX I EM, 10.75 10.88
~DOCU I EM, 10.89 10.95 ITlME I EM, I I ISTOR I EM, I I
BKC 2/23/97 CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
Calibrated schedule constant, A: 2.66 (apriori value was 3.0)
BKC 2/23/97 CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
Presentation Outline: Model Calibration
Calibration Procedures I
COCOMO I1 1997 Model Parameters
Accuracy Results
USC COCOMO Software Status
COSTAR Software Status
Calibration Data Status
BKC 2/23/97 CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
Accuracy Results: Forecast accuracy measured with Proportional Error (PE) :
[pn/r,y, + p ~ ~ c t ] - 1, (PMe,yt - PMact) 2 0 - [PM,, + PM,, ] + 1, ( PMest - PMact ) < 0
I Effort I Before Stratification ( After Stratification
I Schedule ( Before Stratification I After Stratification
BKC 2/23/97 CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
Effort Proportional Error before Stratification
BKC 2/23/97 CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
Effort Proportional Error after Stratification
BKC 2J23197
Organization Number
CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
USC COCOMO Software Status: There is a initial version available for MS Windows, Sun OS, and Java
+ Has new calibrated values
+ Confidence ranges (optimistic, most likely, pessimistic)
+ User definable Cost Drivers: USR1, USR2
+ Schedule input is now project wide
+ New reference manual
+ New values can be manually input for all cost drivers
+ Version changed to COCOMO II.199Y .X (where Y is the year number and X is the version within that
BKC 2/23/97 CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
USC COCOMO Future Work:
BKC 2l23197 CENTER FOR SOFTWARE ENGINEERING
Entry of actuals for periodic tracking of project and data submission
Calibration of constant and exponent
Incremental ratings between Very Low, Low, Nominal, High, Very High, Extra High
Text entry for SU, AA, UNFM
New Help file
UNIVERSITY OF SOUTHERN CALIFORNIA
COSTAR Software Status Commercialized version of COCOMO
Beta version of COCOMO I1 model available
New values will be put in the model soon
BKC 2/23/97 CENTER FOR SOFTWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
Calibration Data Status More project data is required to facilitate better calibration of the general COCOMO I1 Post-Architecture model
We hope the use of USC COCOMO software will facilitate collection and submission of data
If you calibrate the model to your local organization (constant and exponent) - we would like to have your observations in our repository to be used for full model calibration
We plan to make annual updates to the cost driver values and release them on a regular cycle.
BKC 2/23/97 CENTER FOR SOF.TWARE ENGINEERING
UNIVERSITY OF SOUTHERN CALIFORNIA
Information Sources : Phone: 213-740-6470
Email : [email protected]
Web site: http://sunset.usc.edu/COCOMOII/Co~~mo.html
+ Affiliate Prospectus
+ Model Definition Manual (ver. 1.4)
+ Data Collection Form (ver. 1.6)
+ USC COCOMO Software and User's Manual
+ Java COCOMO I
+ Little Expert COCOMO Calculator
BKC a23197 CENTER FOR SOFTWARE ENGINEERING
University of Southern California I C I S I E I Center for Software Engineering L
Outline
COCOMO I1 Status: Brad Clark - Model calibration
- Tool Status - Data Status
-F COCOMO I1 Plans: Barry Boehm - Tool Plans
- Model Plans - Training and data collection plans
Universi~y of Southern California
( c ( S I E ( Center for Software Engineering - Tool Plans: USC COCOMO 11.1997.1
Calibration to an organization's data - Effort andlor schedule
- Coefficient or also exponent
Intermediate rating levels
Updated Madachy risk assessment model 1
Added reuse parameters: SU, AA, UNFM
Calibration.
- Provide a way of capturing and retaining a set of projects.
- Capability of changing C (constant) and E (exponent) from cocomo equation.
- Provide users with 2 ways of using C and E. I
- Standard cocomo values
- Calibrated values
University of Southern California
C I S B E Center for Software Engineering - Model Plans: Affiliate Priorities
14 Activity distribution
13 COTS integration costs
l2 Sizing improvements
5 C~st/schedule/quality tradeoffs
5 Life cycle tradeoff models
A University of Southern California I C I S I E I Center for Software Engineering -
Effort Distribution bv Activitv EffortIFP varies by language level (LL) - But so does effortISLOC! I
Proposed approach - SLOC, LL ==> Effort:
Determine equivalent 3GL SLOC (3 SLOC) via backfiring
Compute effort as F (3 SLOC)
Apply LL stage multipliers to obtain activity distribution, total effort I
- UFP, LL ==> Effort Determine 3 SLOC by backfiring
Compute effort a F(3 SLOC)
Apply LL stage multipliers to obtain activity distribution, total effort
University of Southern California Center for Software Engineering
4GL Cost and Schedule Effects
1
+ Correspondence school information system + Estimated size: 15 KDSl ALL [4GL], 95 KDSl
COBOL + Actual size: 13.9 KDSl ALL, 93.6 KDSl equiv.
COBOL + Data on phase distribution of effort and
schedule I
University of Southern California Center for Software Engineering
4GL Estimates vs. Actuals
Quantity
Size Schedule Effort Plans&Rqts Prel Design DD/CUT/J&T
COCOMO- COCOMO- COBOL 4GL
Recom. Actual Use
13.9 12.0 4GLA1.6 61.6 Mix:62.7 14.9 COBOL 20.5 3*4GL
26.2 4GL
University of Southern California Center for Software Engineering
Effort Distribution Relative to 3GL Development
Rapid APP* Devel.
LCO, LCA SAT
Spiral-type Ev. Dev., Spiral
I
LLICa LCA 1
SAT
Waterfall, W'fall, IncDev, EvDev,
SYS Spiral-type Devel
I
Spiral-type
I
Spiral, Design-to-Cost, etc.
Det. Design Code, Integ., Test
University of Southern California c I S I E ~ Center for Software Engineering
Proposed UML-Based Sizing Model
Rational' s Universal Modeling Language (B ooch, Jacobson, 'Rumbaugh) approaching de-facto OOD standard
A UML-based early sizing metric would address two major current short falls with Funciton Points - Automated counting
I
- Object-orientation
Rational (Walker Royce) is interested in pursuing such a project
Would need some Affiliates to provide data
Model Elements Class
0 A set of objects that share a common structure and a common behavior
Use caselcollaboration 0 A named behavior involving the collaboration of a society of
objects
Stateloperation 0 The condition of an object; an activity
Interface 0 The public part of an object
Thread
n An active class, capable of concurrent activity with other active classes
Model Elements (cont) Component
a A reusable part, typically having both logicalas well as physical aspects
Node .. .
I 3 A hardware device upon which software may reside andlor execute
Package A container of elements
Note , . A comment, explanation, or annotation
Diagrams Class diagram State machine diagram Sequence diagram Collaboration diagram Activity diagram Use case diagram Component diagram Deployment diagram
Architecture
Enduser Programmers Functionality Softwarn management
I - Logical View Component
AnalysMesters Use Case Behavior
Concurrency I I Dsployment I " i i v i w I System Integrators Performance Scalability Throughput
System Engineering srstem *polow
Delivery, installation Communication
A University of Southern California
c [ S I E I Center for Software Engineering - Training and Data Collection Plans
USC COCOMO 11.199'7.1 or COSTARICALICO as data colleciton instruments
Proposed role for Don Reifer - Meet broader need for COCOMO I1 training
- Make data collection easier, *more efficient
ROLE FOR DON REIFER
Assist team in calibrating COCOMO-I1 - Use his databases, when applicable, to help
calibrate the model
- Run statistical tools to generate goodness of fit and other meaninghl measures
Provide ideas based on his extensive cost modeling experience
Develop public training and help get data for the model's continued refinement
WIN-WIN SITUATION USC and Affiliates get: - Calibration data and the results of an analysis
of about 500 projects
- Extensive knowledge base of experience of one of the leaders in field of parametric modeling
- Public courseware when they need it
Reifer Consultants (RCI) gets: - In-depth knowledge of COCOMO-I1
- Ability to consult and market COCOMO-I1 courseware and make a profit
USCIRCI AGREEMENTS
RCI will not market COCOMO packages - Their focus will be training and consulting
RCI will not market competing packages - Their agreement with Resource
has been terminated
- They have elected to discontinue their SoftCost product line in the
Calculations
support for future
USC will cooperate and get data and .
feedback from the trained model users