Recent Results in Software Process Modeling
Ray Madachy, Ph.D.
C-bridge Internet Solutions
University of Southern California Center for Software Engineering
rmadachy@c-bridge.com, madachy@ usc.edu
1 Introduction
Understanding software process interactions
and feedback is increasingly important given changing
software development and evolution paradigms.
Towards this end, a new graduate course in Software
Process Modeling was developed by this author at the
University of Southern California Center for Software
Engineering (USC-CSE) [1]. It was first offered in the
Fall of 1999, and the term projects were original
investigations into critical process issues.
This abstract summarizes the student research
projects, and highlights some results in terms of
identifying important feedback in software processes.
Rather than focus on previous contributions [2], [3], only
new and previously unreported student work is described
herein. The in-process book Software Process Dynamics
[4] was the primary text for the class. Some of the
student work incorporates new concepts in the book,
such as inter-phase iterative feedback, process
concurrence, personnel factors, learning feedback and
model calibration techniques.
2 Course Summary
The course overviews the field of software
process modeling, and addresses current research issues
with student simulation projects. It is designed for
students and software engineering professionals who are
interested in understanding the dynamics of software
development and assessing process strategies. Process
modeling techniques for both continuous systems and
discrete systems are covered, with a concentration in
system dynamics modeling (continuous systems).
Examples of process and project dynamics
covered are Rapid Application Development (RAD), the
effects of schedule pressure, experience, global feedback
and evolution, work methods such as reviews and quality
assurance activities, task underestimation, bureaucratic
delays, demotivating events, process concurrence, other
socio-technical phenomena and the feedback therein.
These complex and interacting process effects are
modeled with system dynamics using continuous
quantities interconnected in loops of information
feedback and circular causality.
The course demonstrates how knowledge of the
interrelated technical and social factors coupled with
simulation tools can provide a means for software
process improvement. More information can be found in
(1).
3 Simulation Term Projects
Students were instructed in a fairly rigorous
process for developing their models, starting with
definitive statements of modeling goals and identifying
reference behavior. Students were also asked to
interview seasoned experts in their respective areas,
develop surveys as appropriate, and pointers to real
world data were provided. Group reviews and extensive
validation tests were performed, and a standard format
was used for the reports.
Students were allowed to do individual projects
or have teams of two. Several of the projects employed
teams. The five term projects investigated the following
issues:
« the dynamics of architecture development in the
inception and elaboration phases of the Model-Based
Architecting and Software Engineering (MBASE)
process
* COTS glue code development and integration
dynamics
* reuse and language-level effects in software
development
* CMM-based Software Process Improvement (SPI)
strategies
* application of RAD techniques to pre-IPO Internet
companies.
Table 1 categorizes the projects in terms of their high-
level scope and purpose.
Table 1: Project Characterizations
Scope / Portion of | Development | Long-term
Purpose lifecycle project organization
Planning and | MBASE Reuse and CMM
control Architecting | High Level Software
Process Languages Process ‘
‘. improvement
improvement Internet RAD
and
technology COTS Glue
adoption Code
Subsequent sections provide more highlights on the
projects. The more in-depth studies performed by two-
person teams are described first. Very few figures are
included here due to limited space. In order to best
visualize the feedback loops in the models, the reader is
encouraged to read the original reports at [1].
3.1 MBASE Architecting
MBASE is an integrated software development approach
developed at USC-CSE. It is used in departmental
Software Engineering courses. The overall goals of the
MBASE architecting study were to:
* — investigate the dynamics of architecture
development during early MBASE lifecyle phases
* identify the nature of process concurrence in early
MBASE phases
* understand the impact of collaboration and
prototyping on lifecycle parameters.
Some features of the model include:
* — schedule as independent variable
iterative process structures
sequentiality and concurrency of activities
phases - requirements and architecture/design
activities - initial completion, coordination, quality
assurance, iteration
* demand-based resource allocation
* — external and internal precedence constraints
* calibration to CS577A project data (CS577A is a
graduate course in Software Engineering taught by Dr.
Barry Boehm where teams develop multimedia
applications for internal USC library usage).
A high-level overview of the lifecycle artifact flows and
feedback is shown in Figure 1, and the major causal
loops are shown in Figure 2.
Architecture &
Design
Requirements
LEGEND
> Prout of phase
mint > Retum Erors
[7] devetopment pase
Figure 2: MBASE Architecting Causal Loop Diagram
This study fleshed out various prototyping feedback
effects, inter-phase iteration feedback, and the model
was calibrated to actual MBA SE project data.
3.2 COTS Glue Code
This study investigated the overall lifecycle of
incorporating COTS into a product, with a primary focus
on the integration dynamics. Overall goals were to:
* understand glue code development, the COTS
integration process and their correlation
* determine efficient starting points of glue code
development and COTS integration
* calibrate the component parameters from COCOTS
* analyze the impact of new parameters such as ratio
of new and updated COTS component and number of
COTS component.
Figure 3 shows some of the feedback interactions in the
model.
Completed!
Glue Code ae
a
fo _ 7 \
i Application Se
Integrated
System
Applicstion/Glue
J Code Integration
Rate ea
Glue Code Dev, *
—_
SN__ Application
Dev Rate
Application ~,
Design Fats ‘
Glue Code J
Figure 3: COT Glue Code Causal Loop Diagram
This study produced valuable results in terms of COTS
and glue code integration planning.
3.3, Reuse and High Level Language
This study looked at dynamics of reuse as well as the
effects of employing high-level languages. Goals for the
project included:
* — investigating project reuse dynamics
* — productivity and effort of individual phases
* understand the effects of different language levels.
Some features of the model are
+ rework is included
«learning curve formulations
* — increased training for higher level languages.
See Figure 4 for the primary effects in the model. This
work produced a validated model containing reuse and
language effects, including a rate-based representation of
reuse impact on project size that can serve as an
archetype structure.
¢ Complete Tmkr ,
x4
Reuse Tasks
‘
Design Approve Task
«oY
» “4
Cade Adoption
*
sPecple —— Predacevi
a4
yefnct Dersity
.
’
Cotnerasication
roerbeud
,
Exes Lesting Curve %
Langsnge Level
Figure 4: Reuse and High Level Language Causal Loop
Diagram
3.4 CMM-Based Software Process Improvement
This work investigated software process improvement
based on the CMM. Subgoals included the following:
* provide insight into complex process behavior
help evaluate different approaches for improvement
support planning, tracking and prediction
reduce costs
reduce cycle time
reduce defects
An existing model [5] was adapted for use at Xerox. It
was based on the scenario of a X erox development group
working from just assessed as a Level 2 organization
moving towards achieving Level 3. Figure 5 shows the
model at a high level, including feedback loops.
Pros Make Minor SPI
Suggestions
New Hire
Attitudinal
Type and Retains
Experienced
Pros
Figure 5: SPI Model Feedback Relationships
Referring to the figure, the Lifecycle process
models how software size, effort, quality, and schedule
relate to each other in order to produce a product. SPI
benefits are modeled as percent reductions in either size,
effort, error rate or schedule.
In People, three attitudes of staff that affected
the potential benefit of process improvement: pro-SPI
people, con-SPI people, and no-care people. The
attitudinal mix and the pro/con ratio can affect the
overall potential benefit realized by a SPI effort.
KPA Processing models the timing of the flow
of process improvements into the lifecycle and people
subsystems.
The following causal loop description is the basis of the
simulation model.
Major software process improvement (SPI) efforts
are piloted and deployed based on project cycle time
(i.e., pilot on one project, tailor and deploy on the
subsequent project).
2. Major SPIs increase maturity (the probability of
successfully achieving your goals).
3. Increased maturity attracts new hires and retains
experienced staff that are “pro SPI” (i.e., they
support and participate in SPI activities and are
attracted to success and innovation).
4. Pro-SPI staff make minor SPI suggestions.
5. Major and minor SPIs decrease cycle time.
6. Decreased cycle time enables more major and minor
SPIs to be accomplished.
7. Go back to 1 and repeat the cycle.
This model was calibrated for the X erox environment,
and the results are being used for internal SPI planning.
3.5 Internet RAD
The goals for this project included:
* investigating the dynamics of pre-IPO Internet
companies
* contrasting the dynamics to non-Internet software
development
* surveying companies and determine major impact
factors.
Some features of the Internet RAD model are:
* a modified evolutionary delivery lifecycle with
small teams
schedule minimization.
outsourcing considerations
defect detection and elimination
short term and long-term feedback
Internet preview and web-site personalization
+ model sectors - Specification and Design,
Outsourcing, Development, Integration and
Personalization, Human Resources.
Short-term and long-term feedback to the web site
integration can be seen in the Integration and
Personalization sector in Figure 6.
Figure 6: Internet RAD Integration and Personalization
Sector
Contributions of this study include the lifecycle
definition for Internet-unique instant software delivery,
and identification of the two inherent types of feedback.
4 Feedback Summary
A summary of the most important feedback
effects from these studies are shown in Table 2. Refer to
the original papers for specific attributes of the feedback.
Table 2: Summary of Major Feedback
Project Major Feedback Effects
MBASE Feedback from prototype
Architecting evaluation that impacts the
requirements and architecture.
Inter-phase iterative feedback
from approval activities, reviews
and testing.
inforcing feedback from
initially completed items.
Reinforcing feedback from
prototyping to iteration rate.
COTS Glue Code Connected feedback loops
between application
development, glue code
development and system
integration.
Reuse and High
Level Languages
Learning feedback effects on
development and reuse rates.
Professor's note: Not included is
the important feedback between
the newly developed and
reusable software, and changing
specifications which contribute
to reuse growth and decay
Software Process
Improvement (SPI)
Major feedback loops between
process improvements, process
maturity, personnel mix and
project cycle time.
Internet RAD Early error detection and rework.
Short-term "instant" feedback on
bugs from the internet site.
Long-term customer feedback on
desired changes and capabilities
after "personalization" of the
site.
References
[1] CS599 Software Process Modeling main web page
(includes full project reports and other system dynamics links),
http://sunset.usc.edu/classes/cs599_99
[2] Madachy R, A Software Project Dynamics Model for
Process Cost, Schedule and Risk Assessment, Ph.D.
Dissertation, Dept. of Industrial and Systems Engineering,
University of Southern California, December 1994
[3] Madachy R, Tutorial: Cost/Schedule/Process Modeling via
System Dynamics, Proceedings of the 14th International Forum
on COCOMO and Software Cost Modeling, Los Angeles, CA,
October 1999
[4] Forthcoming book: Madachy R, Boehm B, Software
Process Dynamics, IEEE Computer Society Press,
Washington, D.C., 2001,
http://sunset.usc.edu/Research_Group/ray/spd.
[5] Burke S, Radical Improvements Require Radical Actions:
Simulating a High-Maturity Software Organization, CMU/SEL
96-TR-024, 1997
CC BY-NC-SA 4.0