Transcript Jeff Poulin

Cost Drivers Affecting
Deployment of COTS-Based
Systems
Jeffrey Poulin and Walt Johnson
Jeffrey.Poulin,W.A.Johnson @ lmco.com
(607)751-6899
Lockheed Martin Systems Integration
Owego, NY
Agenda
•
•
•
•
•
Overview of COTS programs
Cost drivers used in sizing model
Requirements history
Impact of cost drivers and requirements
Lessons learned
Program A
• Purpose: Migrate data from 100+ applications into a data warehouse
for common data mining.
• Size from Cost Model:
– 4375 Function Points
– 3.5m SLOC COBOL
– 50k SLOC Java/other
• Technologies used:
–
–
–
–
–
IBM System 390/MVS
DB2 Database
Datalogic GUI Builder
PRISM code generator for COBOL
Various APIs
Cost Drivers for Program A
Cost Drivers
Initial Warehouse, "y" or "n"
Planned
y
Actual
y
Number of tables
Number of sources
Number of users at acceptance
Number of users at implementation (less those at acceptance)
142
15
50
300
156
100+
50
300
Number of static reports
120
120
25
25
1
2
1
2
1
2
1
2
y
y
Number of people on the development team
Next Three items use the average of the group for the following scale:
0 for expert, 1 for k nowledgeable, 2 for some experience, 3 for untrained
Level of Development Team's knowledge on Database
Level of Development Team's knowledge on UNIX
Level of Development Team's knowledge on GUI
Level of Development Team's knowledge on Warehouse Concepts
Do a proof-of-concept pilot warehouse, "y" or "n"
Requirements Volatility for Program A
• Change Requests
– Release 1: 61
– Release 2: 34
– “Not a large cost driver” (Managed by stable Configuration
Management Process)
• Business Rules:
– 15 planned
– 100+ actual
– Caused multiple design iterations of DB Schema
Cost Impact to Program A
• Cost impact as a function of planned vs. actuals:
– Release 1: 5.1x
– Release 2: 2.2x
– Overall: 2.6x
• Lessons Learned:
– Insist on complete documentation of business rules and legacy
data.
– Diverse nature of interfaces demanded custom handling and
limited reuse.
Program B
• Purpose: Develop and integrate products for processing addresses on
mail pieces using optical character recognition
• Size from Cost Model:
– 158k SLOC of development
– 180k SLOC of reuse
– 10 major suppliers of COTS components
• Technologies used:
–
–
–
–
–
IBM Risc System/6000 with AIX
Oracle Database
Visual Studio
Java/J-Builder GUI
Various APIs
Cost Drivers for Program B
• Total System Solution- 55%
–
–
–
–
–
System Design
Software development
COTS licenses
Integration of COTS and existing components (black-box)
Hardware
• Network- 13%
• Project management- 10%
• Interface development- 7%
– To all legacy systems
• Training – 1%
• Support- 13%
– Maintenance
– Enhancements
Requirements Volatility for Program B
• Requirements
– June 1998: 543
– Aug 1999: 1208
– Feb 2000: 1186
July 1999: 934
Oct 1999: 1299
Dec 2000: 963
Total Requirements
1400
1200
1000
800
600
400
200
0
June'98
July'99
Aug'99
• 41.8% Fluctuation
Oct'99
Feb'00
Dec'00
Pr
ev
8/ iou
9/ s
8/ 2 00
23 0
9/ /20
24 00
10 /20
/2 00
5
11 /20
/8 00
11 /20
/1 00
12 5/2
/1 00
2/ 0
2
1/ 000
4
1/ /2 0
12 01
/2
00
1
Open vs. Closed CRs
Change Requests for Program B
CR Status
60
50
40
30
20
10
0
Week Of
• 5.1% CR rate
• Not a major impact
SLOC History of Program B
k SLOC
Total SLOC by Release
400
350
300
250
200
150
100
50
0
Reused
New/Modified
Release
1
Release Release Release
2
3 (est)
4 (est)
Cost
Model
Estimate
• High correlation between cost model prediction and EAC
• Inability to maintain COTS-based vision
Cost Impact to Program B
• Cost impact based on Estimate at Complete:
– Release 1: 1.05x
– Release 2: 1.11x
• Lessons Learned:
– Necessity of managing customer expectation in COTS-based
environment
– Requirement to adapt customer business processes to obtain vision
of using COTS-based solutions (very difficult!)