USC
C S E
University of Southern California
Center for Software Engineering
CORADMO in 2001:
A RAD Odyssey
Cyrus Fakharzadeh
[email protected]
16th International Forum on COCOMO and
Software Cost Modeling
1
USC
C S E
University of Southern California
Center for Software Engineering
Introduction
RAD (Rapid Application Development)
• an application of any of a number of
techniques or strategies to reduce
software development cycle time
CORADMO
• COCOMO II model extension
• Focuses on software development schedules
and costs using RAD techniques
2
USC
C S E
University of Southern California
Center for Software Engineering
Constructive Rapid Application
Development Model
• Calculates/predicts
– schedule (months, M)
– personnel (P)
– adjusted effort (person-months, PM)
• Based on
– Effort and schedule distribution to the various phases
– Selected schedule driver ratings impacts on the M, P,
and PM of each phase.
3
USC
C S E
University of Southern California
Center for Software Engineering
Six Classes of Strategies for RAD
•
•
•
•
•
•
Reuse, Very High-level Languages (RVHL)
Development Process Reengineering (DPRS)
Collaboration Support (CLAB)
Architecture, Risk Resolution (RESL)
Prepositioning Assets (PPOS)
RAD Capability of Personnel (RCAP)
4
USC
C S E
University of Southern California
Center for Software Engineering
RAD Opportunity Tree
B usiness p ro cess reeng ineering - O
D evelo p m ent p ro cess reeng ineering - D PR S
E lim inating Tasks
R educing Tim e P er Task
R educing S ingle-P oint Failure R isks
R eusing assets - R VH L
Ap p licatio ns g eneratio n - R VH L
D esig n-to -sched ule - O
T o o ls and auto m atio n - O
W o rk stream lining (80-20) - O
Increasing p arallelism - R ESL
R ed ucing failures - R ESL
R ed ucing their effects - R ESL
Early erro r elim inatio n - R ESL
R educing B acktracking
Pro cess ancho r p o ints - R ESL
Im p ro ving p ro cess m aturity - O
C o llab o ratio n sup p o rt - C L AB
Activity N etw ork S tream lining
M inim izing task d ep end encies - D PR S
Avo id ing hig h fan-in, fan-o ut - D PR S
R ed ucing task variance - D PR S
R em o ving tasks fro m critical p ath - D PR S
Increasing E ffective W orkw eek
B etter P eople and Incentives
Transition to Learning O rganization
Prep o sitio ning reso urces - PPO S
N ig htly b uild s, testing - PPO S
W eekend w arrio rs, 24x7 d evelo p m ent - PPO S
M o re R AD exp erience - R C AP
-O
O : co vered b y classic cu b e ro o t m o d el
5
USC
C S E
University of Southern California
Center for Software Engineering
Background
COCOMO II Schedule shortfalls:
• Reflects projects optimized for minimum cost
• Model does not address RAD strategies
COCOMO II.2000 Duration Calculation
Cube Root Law: Months ~ 3.67 (Person-Months)f
where 0.28 ≤ f ≤ 0.34
CORADMO differs from COCOMO:
• A square root instead in computing the number of months
needed to complete a small project
• Square root law (i.e. f = 0.5)
6
USC
C S E
University of Southern California
Center for Software Engineering
COPSEMO
Constructive Phased Schedule and Effort Model
Inputs: the baseline effort and schedule from
COCOMO II
Outputs: the effort and schedule by phase
needed for CORADMO.
Phases: Inception, Elaboration, Construction,
and Transition
Source: MBASE/RUP (Model-Based
Architecting & Software Engineering/Rational
Unified Process) life-cycle model
7
USC
University of Southern California
Center for Software Engineering
COPSEMO Months as F(PM)
Previous
RAD/COCOMO
bridge for large
projects
20.00
16.00
Months (M)
C S E
~3*cube-root
COCOMO II
(minumum cost)
12.00
Square root
(minimum schedule)
small projects
8.00
4.00
0.00
0
16
32
48
64
80
96
112 128 144
Effort (PM)
CII-M [Cube Root]
Original COPSEMO-M
New COPSEMO-M [Square Root]
8
USC
C S E
University of Southern California
Center for Software Engineering
Physical Model
RVHL
RESL
DPRS
CLAB
COCOMO II
cost drivers
COCOMO II.2000
via
RESL; Baseline
COCOMO_charts.xls
effort, schedule
Baseline Effort & Sched.
Phase
Distributions
(COPSEMO Extension)
Schedule calculated; SCED removed;
PM & M distributed per phase
Effort, schedule
PPOS
RCAP
RAD
Extension
(CoRADMo.xls)
by phase;
No SCED
RAD effort,
schedule
by phase
9
USC
C S E
University of Southern California
Center for Software Engineering
Results
• Delphi Exercise Forms distributed
• Experts from Academia, Industry and Government
– Affiliates, Professors, and Researchers
• EMR (Effort Multiplier Range)
– Highest divided by Lowest across the rating scale for
effort
• SMR (Schedule Multiplier Range)
– Highest divided by Lowest across the rating scale.
10
USC
C S E
University of Southern California
Center for Software Engineering
% Effort per phase
Original Delphi
Mean
Inception – I
6.0
10.29
Delphi
Standard
Deviation
4.75
Elaboration – E
24.0
23.71
5.38
Construction – C
76.0
71.29
12.00
Total I, E, & C
106.0
105.29
4.24
11
USC
C S E
University of Southern California
Center for Software Engineering
% Schedule per phase
Original Delphi
Mean
Inception – I
12.5
15.71
Delphi
Standard
Deviation
4.99
Elaboration – E
37.5
29.86
5.64
Construction – C
62.5
63.14
11.94
Total I, E, & C
112.5
108.71
6.94
12
USC
C S E
University of Southern California
Center for Software Engineering
Reuse, Very High-level Languages
Degree to which re-use of artifacts other than code
and/or very high-level languages are utilized
EMR
RVHL
SMR
Delphi
Delphi
Delphi Standard
Delphi Standard
Original
Original
Mean
Mean
Deviation
Deviation
Inception
Elaboration
Construction
1.16
1.25
0.08
1.16
1.22
0.06
1.07
1.25
0.09
1.07
1.24
0.08
1.00
1.16
0.11
1.00
1.13
0.09
13
USC
C S E
University of Southern California
Center for Software Engineering
Development Process Reengineering
Measures the degree to which the project and
organization allow and encourage streamlined or
reengineered development processes
EMR
DPRS
SMR
Delphi
Delphi Standard
Delphi
Original
Original
Mean
Mean
Deviation
Inception
Elaboration
Construction
Delphi
Standard
Deviation
1.33
1.32
0.02
1.33
1.25
0.05
1.21
1.24
0.04
1.21
1.21
0.02
1.21
1.30
0.06
1.21
1.22
0.04
14
USC
C S E
University of Southern California
Center for Software Engineering
Collaboration Support
Accounts for Multisite tool support plus special
collaboration tools, yields a reduced effect on
schedule and effort
EMR
CLAB
SMR
Delphi
Delphi Standard
Delphi
Original
Original
Mean
Mean
Deviation
Inception
Elaboration
Construction
Delphi
Standard
Deviation
1.51
1.34
0.09
1.51
1.27
0.13
1.34
1.23
0.07
1.34
1.23
0.06
1.18
1.23
0.08
1.18
1.21
0.05
15
USC
C S E
University of Southern California
Center for Software Engineering
Architecture, Risk Resolution
Same as COCOMO II RESL
EMR
RESL
SMR
Delphi
Delphi Standard
Delphi
Original
Original
Mean
Mean
Deviation
Inception
Elaboration
Construction
Delphi
Standard
Deviation
1.00
1.24
0.34
1.00
1.21
0.35
1.00
1.24
0.34
1.00
1.23
0.35
1.00
1.27
0.33
1.33
1.35
0.29
16
USC
C S E
University of Southern California
Center for Software Engineering
Prepositioning Assets
Degree to which assets are pre-tailored to a project
and furnished to the project for use on demand
EMR
PPOS
SMR
Delphi
Delphi Standard
Delphi
Original
Original
Mean
Mean
Deviation
Inception
Elaboration
Construction
Delphi
Standard
Deviation
1.10
1.13
0.02
1.25
1.22
0.02
1.10
1.14
0.04
1.25
1.23
0.02
1.10
1.20
0.04
1.25
1.26
0.03
17
USC
C S E
University of Southern California
Center for Software Engineering
RAD Capability of Personnel
Accounts for the effects of RAD personnel capability
& experience in RAD projects
EMR
RCAP
SMR
Delphi
Delphi Standard
Delphi
Original
Original
Mean
Mean
Deviation
Inception
Elaboration
Construction
Delphi
Standard
Deviation
1.50
1.48
0.15
3.00
2.48
0.26
1.50
1.44
0.14
3.00
2.48
0.25
1.50
1.46
0.14
3.00
2.49
0.25
18
USC
C S E
University of Southern California
Center for Software Engineering
Example
With RCAP = Nominal => PM=25, M=5, P=5
Result: The square root law: 5 people for 5 months: 25 PM
With RCAP=XH (Extra High) => PM=20, M=2.8, P=7.1
Result: A super team can put on 7 people and finish in 2.8
months: 20 PM
With RCAP = XL (Extra Low) => PM=30, M=7, P=4
Result: Trying to do RAD with an unqualified team makes them
less efficient (30 PM)
19
USC
C S E
University of Southern California
Center for Software Engineering
RCAP Effort/Schedule Effect
16
14
12
M
10
8
RCAP = XL
6
4
RCAP = XH
2
0
0
10
20
30
40
50
PM
3.7*(Cube root)
3*(Cube root)
Square root
20
USC
C S E
University of Southern California
Center for Software Engineering
Next Steps
• Complete another Delphi Round
• Gather more RAD data
¾Please contact me if you have some
• Analyze Data from RAD projects
• Bayesian Analysis
• Calibrate Model
21
USC
C S E
University of Southern California
Center for Software Engineering
Issues for Breakout Group
• Treatment of square root, cube root
models
• Treatment of RAD drivers
• Relevance to your RAD experience
• Expediting data collection
22