A Dynamic Integrated Framework for Software Process Improvement

Abstract

Current software process models (CMM, SPICE, etc.) strongly recommend the application of statistical control and measure guides to define, implement, and evaluate the effects of different process improvements. However, whilst quantitative modeling has been widely used in other fields, it has not been considered enough in the field of software process improvement. During the last decade software process simulation has been used to address a wide diversity of management problems. Some of these problems are related to strategic management, technology adoption, understanding, training and learning, and risk management, among others. In this work a dynamic integrated framework for software process improvement is presented. This framework combines traditional estimation models with an intensive utilization of dynamic simulation models of the software process. The aim of this framework is to support a qualitative and quantitative assessment for software process improvement and decision making to achieve a higher software development process capability according to the Capability Maturity Model. The concepts underlying this framework have been implemented in a software process improvement tool that has been used in a local software organization. The results obtained and the lessons learned are also presented in this paper.

Software Quality Journal, 10, 181194, 2002

2002 Kluwer Academic Publishers, Manufactured in The Netherlands.MERCEDES RUIZ [email protected] of Computer Languages and Systems, Escuela Superior de Ingeniera,

University of Cdiz, SpainISABEL RAMOS AND MIGUEL TORO {isabel.ramos, mtoro}@lsi.us.esDepartment of Computer Languages and Systems, Escuela Tcnica Superior de Ingeniera Informtica

University of Seville, SpainAbstract. Current software process models (CMM, SPICE, etc.) strongly recommend the application

of statistical control and measure guides to dene, implement, and evaluate the effects of different

process improvements. However, whilst quantitative modeling has been widely used in other elds, it

has not been considered enough in the eld of software process improvement. During the last decade

software process simulation has been used to address a wide diversity of management problems. Some

of these problems are related to strategic management, technology adoption, understanding, training

and learning, and risk management, among others. In this work a dynamic integrated framework for

software processimprovement ispresented. Thisframework combinestraditional estimation models

with an intensive utilization of dynamic simulation models of the software process. The aim of this

framework is to support a qualitative and quantitative assessment for software process improvement and

decision making to achieve a higher software development process capability according to the Capability

Maturity Model. The conceptsunderlying thisframework have been implemented in a software process

improvement tool that has been used in a local software organization. The results obtained and the

lessons learned are also presented in this paper.Keywords: software process modelling and simulation, process improvement, process maturity, dynamic

models1. IntroductionOver the past few decades software complexity has signicantly increased in such

a way that software has replaced hardware as having the principal responsibility

for much of the functionality provided by current systems. This increasing role of

software, the problems related to cost and schedule overruns, and the customer

perception of low product quality have changed the focusof attention towards

the maturity of software development practices. Although the software industry

hasreceived signicant help by meansof Computer Aided Software Engineering

(CASE) tools, new programming languages and approaches, and more advanced

and complex machines, there isa lack of processanalysistoolsfor organizations

interested in improving their process performance.Dynamic modeling and simulation asprocessimprovement toolshave been intensively used in the manufacturing area. Currently, software process modeling andA Dynamic Integrated Framework for Software

Process Improvement182 RUIZ, RAMOS, AND TOROsimulation are gaining an increasing interest among researchers and practitioners

as an approach to analyze complex business and solve policy questions.In previouswork (Ruiz, Ramos, and Toro, 2001) we attempted to apply a complex

dynamic model to support process improvement in a local software development

organization. The non-existence of a historical database and of measurement practices inside this organization made it impossible, as there were no numerical drivers

to supply the model parameters and functions. Then, we decided to apply Eberleins

work (Eberlein, 1989) about understanding and simplication of models to obtain

a reduced dynamic model capable of reproducing the software process dynamics,

yet with less initial information required. But simulation is only effective if both

the model and the data used to drive it accurately reect the real world. Thus, the

construction of the model itself points to what metric data must be collected and

helpsasa clear guideline on what to collect.In thispaper an approach isproposed that combinestraditional estimation techniqueswith System Dynamicsmodeling. The aim of thiscombination isto build a

framework to support a qualitative and quantitative assessment for software process improvement and decision making. The purpose of this dynamic framework

isto help organizationsto achieve a higher software development processcapability according to the Capability Maturity Model (Paulk et al., 1993). The dynamic

modelsbuilt inside thisframework provide the capability of gaining insight over the

whole life cycle at different levels of abstraction. The level of abstraction used in a

certain organization will depend on itsmaturity level. For instance, in a level 1 organization the simulator can establish a baseline according to traditional estimation

models from an initial estimate of the size of the project. With this baseline, the

software manager can analyze the results obtained with the simulation of different

process improvements and study the outcomes of over or underestimating cost or

schedule. During the simulation metric data are saved. These data conform to the

SEI core measures (Carleton et al., 1992) recommendation, and are mainly related

to cost, schedule and quality.The structure of the paper isasfollows. Section 2 providesa brief overview of the

work conducted in the eld of software process simulation. In Section 3, the justication found to develop the integrated framework is presented. Section 4 describes

in detail, the fundamental basis and structure of this framework. The implementation and results obtained when applying it inside a local organization are discussed

in Section 5. Finally, Section 6 summarizesthe paper and drawsthe conclusions

and lessons learnt.2. Software process simulationSimulation can be applied in many critical areas in support of software engineering.

It enables one to address issues before these issues become problems. Simulation is

more than just a technology, asit forcesone to think in global termsabout system

behavior, and about the fact that systems are more than the sum of their components (Christie, 1999). A simulation model is a computational model that representsFRAMEWORK FOR SOFTWARE PROCESS IMPROVEMENT 183an abstraction or a simplied representation of a complex dynamic system. Simulation modelsoffer, asa main advantage, the possibility of experimenting with different management decisions. Thus, it becomes possible to analyze the effect of those

decisions in systems where the cost or risks of experimentation make it unfeasible.

Another important factor isthat simulation providesinsightsinto complex process

behavior which is not possible to analyze by means of stochastic models. Like many

processes, software processes can contain multiple feedback loops, such as those

associated with the correction of defects. Delays resulting from these defects may

range from minutes to years. The resulting complexity makes it almost impossible for mental analysis to predict the consequences. The most frequent sources of

complexity in real software processes are: Uncertainty. Some real processes are characterized by a high degree of uncertainty. Simulation modelsmake it possible to deal with thisuncertainty asthey

can represent it exibly by meansof parametersand functions. Dynamic behavior. Some processes may have a time dependent behavior. There

isno doubt that some software processvariablesvary their behavior asthe time

cycle progresses. With a simulation model it is possible to represent and formalize the structures and causal relationships that dictate the dynamic behavior of

the system. Feedback. In some systems the result of a decision made in a certain moment

can affect their future behavior. For example, in software projects the decision

of reducing the effort assigned to quality assurance activities has different effects

over the whole progress of these projects.Thus, the common objectives of simulation models consist of supplying mechanisms to experiment, predict, learn, and answer questions such as: What if ?

A software process simulation model can be focussed on certain aspects of the

software process or the organization. It is important to bear in mind that a simulation model constitutes an abstraction of the real system, and so it represents only

the parts of the system that have been intended to be modeled. Furthermore, currently available modeling toolssuch asithink[circleR] (High Performance Systems, 2001),

POWER-SIM[circleR] (PowerSim Corporation, 2001), and Vensim[circleR] (Ventana Systems,

2002) help to represent the software development process as a system of differential equations. Thisisa remarkable characteristic, asit makesit possible to formalize and develop a scientic basis for software process modeling and improvement.

Some noticeable applicationsof thisdynamic approach to model software process

can be found in (Kellner, Madachy, and Raffo, 1999).3. JusticationAlthough traditional methodsfor software project management have revealed their

weaknesses during the last decades, there is common agreement that they are still

useful. We think that it isimportant to integrate traditional methodsand process

simulation under a common approach, in order to obtain a valuable tool to design184 RUIZ, RAMOS, AND TOROprocessimprovementsand evaluate their effects. Rodriguesand Bowers(1996),

draw the conclusions obtained after having compared both approaches, and point

out the necessity of integration. Traditional and dynamic approaches have similar

objectives, yet the perspective under which they work is completely different. Traditional methods are normally based on a Top-down decomposition of the software

project, while the dynamic method can be characterized by the aggregation process

it is focussed on, according to which, some features of a project are joined together

under a simulation model. In accordance with what has been said, it can be deduced

that dynamic modelsare suitable to deal with problemsplaced at the strategic level,

while traditional methods are useful at the operational level of software projects.With the development of a Dynamic Integrated Framework for Software Process Improvement (DIFSPI) we offer a methodology and working environment to

join both approachesand to allow project managersand membersof the Software Engineering Improvement Group (SEIG) to design and evaluate new process

improvements. One of the main objectivesof DIFSPI isto support the evolution of

the maturity level of an organization according to the Capability Maturity Model

(Paulk et al., 1993). The process of design and development of both the framework

and the dynamic modelsthat integrate it, allowsone to dene a metricscollection

program. These metrics are necessary to both initialize and validate the dynamic

models. This metrics collection is not only useful for the dynamic models; it also

servesasan invaluable opportunity to obtain a real knowledge of the state of the

software processes inside an organization. This knowledge is essential before tackling any process improvement.4. DIFSPI development4.1. Conceptual approachUsing simulation for process improvement in conjunction with CMM is not a new

idea. As a matter of fact, (Christie, 1999) suggests that CMM is an excellent incremental framework to gain experience through process simulation. Nevertheless,

there is a lack of a dynamic framework capable of assessment in the achievement of

higher process maturity. One of the main features of DIFSPI is that this assessment

is provided not only by using the associated nal tool, but during the development

of the whole dynamic framework. The reason for thisisthat the benetsthat can be

obtained with the utilization of dynamic modelsinside an organization, are directly

related to the knowledge and the empirical information the organization hasabout

its processes. Figure 1 illustrates this idea. It shows the existing causal relationships

among the maturity level of the organization, the utilization of dynamic modelsand

the benetsobtained.The positive feedback loop comes to illustrate the causal relationship that reinforcesthe metricscollection inside the organization. The metricscollected will be

used to calibrate and initialize the dynamic models. Lower maturity organizations

are characterized by the absence of metric programs and historical databases. In

this case, it is necessary to begin by identifying the general processes and the information that hasto be collected about them. The questionsof what to collect, atFRAMEWORK FOR SOFTWARE PROCESS IMPROVEMENT 185stim atellectionFigure 1. Causal relationships derived from the development and utilization of dynamic models.what frequency, and with what accuracy have to be answered at this moment. The

design process of dynamic models helps to come to a solution to these questions.

When developing a dynamic model it isrequired to know: a) what isintended to

be modeled, b) the scope of the model, and c) what behaviors need to be analyzed. Once the model isdeveloped, it needsto be initialized with a set of initial

conditions in order to execute the runs and obtain the simulated behaviors. These

initial conditionscustomize the model to the project and to the organization to

be simulated and they are effectively implemented by a set of initial parameters.

These parameters that rule the evolution of the model runs answer precisely the

former question of what data collect: those data required to initialize and validate

the model will be the main componentsof the metricscollection program.Once the componentsof the metricscollection program have been dened it can

be implemented inside the organization. This process will lead to the achievement

of a historical database. The data gathered can then be used to simulate and empirically validate the dynamic model. When the dynamic model hasbeen validated, the

results of its runs can be used to generate a simulated database; with this database it

is possible to perform process improvement analyses. An increase in the complexity

of the actionsintended to be analyzed will directly lead to an increase in the complexity of the dynamic model required and, therefore, to a new metricscollection

program for the new simulation modules.The bottom half of Figure 1 illustratesthe effectsderived from the utilization

of dynamic modelsin the context of processimprovement. Using dynamic modelswhich have been designed and calibrated according to an organizationsdata

provides three important benets. Firstly, the data of the simulation runs can be

used to predict the future evolution of the project. The graphical representations

of these data show the evolution of the project from a set of initial conditions

(which have been established by the initialization parameters). By analyzing these

graphics, organizations with a low level of maturity can obtain a useful qualitative knowledge about the evolution of the project. Asthe maturity level of the

organization increases, the knowledge about its processes is also higher and thePrediction186 RUIZ, RAMOS, AND TOROsimulation runs can be used as real quantitative estimates. These estimates help

to predict the future evolution of the project with an accuracy that isintimately

related to the uncertainty of the initial parameters. Secondly, it becomes possible

to dene and experiment with different processimprovementsby analyzing the different simulation runs. This capability helps in the decision-making process, as only

the improvements which gave the best results will be implemented. Moreover, one

of the most remarkable thingshere isthat these experimentsare performed with

no cost and risk to the organization as they use the simulation of scenarios. Thirdly,

the simulation model can also be used to predict the cost of the project; this cost

can be referred to as the overall cost, or to a hierarchical decomposition of the total

cost, as for instance, the cost of quality or revision activities. These three benets

are the main factorsthat lead to the achievement of a higher maturity level inside

an organization according to CMM.4.2. DIFSPI structureProject management iscomposed of activitieswhich are intimately interrelated in

the sense that a certain action performed over a determined area will possibly affect

other areas. For instance, a time delay will always affect the cost of the project but

it may or may not affect the morale of the development team, or the quality of

the product. The interactionsamong the different areasof project management

are so strong that on some occasions the throughput of one of them can only be

achieved by reducing the throughput of another. A clear example of thisbehavior

can be found in the frequent practice of reducing the quality, or the number of

requirementsto be implemented in a certain version of the product with the aim

of meeting the time or cost estimates.Dynamic modelshelp to understand the integrated nature of project management,

as they describe it through different processes, structures, and main interrelationships. In the framework proposed here, project management is considered as a set

of dynamic interrelated processes. Projects are composed of processes. Each process

iscomposed of a seriesof activitiesdesigned for the achievement of an objective

(Paulk et al., 1993). From a general point of view, it could be said that projects are

composed of processes which fall into one of the following categories: Management process. This category collects all those processes related to the

description, organization, and control of the project. Engineering process. All those processes related to the specication and development activitiesof the software product are collected in thiscategory.Both categoriesinteract during the time cycle of the project asFigure 2 shows.

From an initial plan performed by the project management processes, engineering

processes begin to be executed. Using the information gathered about the progress

of this second group of processes, project management processes determine the

modicationsthat need to be made to the plan in order to achieve the project

objectives. The DIFSPI proposed follows this same classication and is structuredFRAMEWORK FOR SOFTWARE PROCESS IMPROVEMENT 187Figure 2. Classication of processes for software development.to attend to project management and engineering processes. In both levels, the

utilization of dynamic models to simulate real processes and to dene and develop

a historical database, will be the main feature.4.2.1. Engineering processes in the DIFSPI. On thislevel the dynamic modelssimulate the life cycle of the software product. The benets that simulation provides at

thislevel are the following: To build a model it is necessary to improve the knowledge one has about the

software development process, as it is required to establish the limits and scope

of those real behaviors to be modeled and simulated. The parametersrequired by the model and the tableswhich determine itstime

behavior will constitute the main elementsof a metricscollection program to

dene a historical database. The effective application of thismetricsprogram will feed the database. The

historical data gathered will help assess in the validation and calibration of the

model. The dynamic model will nally simulate the software processes with the knowledge and the maturity that the organization hasat the moment. The utilization of the dynamic model allows the establishment of a baseline for

the project, the investigation of possible improvements, and the development of

a historical database which can be fed either by real or simulated data.The dynamic modelsof thislevel at DIFSPI should follow the levelsof visibility

and knowledge of the engineering processes that organizations have at each maturity level. It isobviousthat the complexity of the dynamic model used in level 1

organizationscannot be the same asthat of the modelscapable of simulating the

engineering processes of, for instance, level 4 organizations.4.2.2. Project management processes in the DIFSPI. Management processes are

divided into two main categories: Plan. Thisgroupsthe processesdevoted to the design of the initial plan and

the required modicationswhen the progressreportsindicate the appearance of188 RUIZ, RAMOS, AND TOROproblems. The modelsof thisgroup integrate traditional estimation and planning

techniquestogether with dynamic ones. Control. In thisgroup all the modelsdesigned for the monitoring and tracking

activities are gathered. These models will also have the responsibility of determining the corrective actionsto the project plan. Therefore, the simulation of

processimprovementswill be of enormousimportance.4.3. Elaboration of the dynamic modelsThe approach followed in the construction of the dynamic modelsisbased on two

fundamental principles: The principle of extensibility of dynamic models. According to this principle,

different dynamic modulesare joined to an initial and basic dynamic model.

Thisinitial model modelsthe fundamental behavior of a software project. Each

one of the dynamic modulesmodelseach one of the key processareaswhich

conform the step to evolve to the next level of maturity. These modules can

be either enabled or disabled according to the objectives of the project

manager or the membersof the SEIG. The principle of aggregation/decomposition of tasks according to the level of

abstraction required for the model. Two levels of aggregation/decomposition

are used: Horizontal aggregation/decomposition according to which different sequential

tasks are aggregated into a unique task with a unique schedule. Vertical aggregation/decomposition according to which different and individual, but interrelated and parallel tasks are considered as a unique task with a

unique schedule.The denition of the right level of aggregation and/or decomposition for the tasks

mainly affectsthe modeling of the engineering activitiesand principally dependson

the maturity level of the process intended to be simulated.To dene the initial dynamic model the common feedback loopsamong the software projectswere taken into account. The objective of thisapproach wastrying to

achieve a generic model and avoid modeling certain behaviorsof concrete organizationswhich might limit the exibility of the DIFSPI. To initialize the functionsand

parameters of the initial model, data originating from historical databases collected

in the available literature were used (Putnam, 1992). By replicating some of the

equations of the initial model it is possible to model the progress to higher maturity

levels.Figures 3 and 4 illustrate the former idea of basic modeling and structure replication. The initial model can be used to simulate software projects developed in

organizations progressing to level 2. Figure 3 uses System Dynamics notation to

illustrate the components developed to model the software development activity.

The number of tasksto be developed isdetermined from an initial estimate ofFRAMEWORK FOR SOFTWARE PROCESS IMPROVEMENT 189revision rate INITIAL SIZE QUALITY development rate Accomplished

tasks Revision

pending

tasks Figure 3. Basic dynamic module for software development modeling.the size of the project. These pending tasks become accomplished tasks according

to the development rate. During this process, errors can be committed. Thus, in

accordance to the desired quality objective for the project, the quality rate and the

revision rate are determined. These two rates govern the number of tasks that are

revised. To model the progress to level 3, the model will make use of a horizontal

decomposition, creating asmany substructuresasphasesor activitiesare present in

the task breakdown structure of the project (analysis, design, code and test, in our

case). According to this approach, each time a complete model or some part of it is

replicated, it will be necessary to dene the new xing mechanisms (dynamic modules) for the new structures. These mechanisms effectively implement the principle

of aggregation/decomposition previously mentioned. The replication of structures

also provides the possibility of replicating the modules related to the project management processes. This replication is especially useful for high maturity level organizations, which will be able to establish process improvement practices for each

certain activity of the life cycle.In Figure 4, the componentsof the dynamic model for level 3 organizationsare

shown. Each one of the four boxes labelled with the name of one phase of the

project is, in fact, a complete dynamic module identical to that shown in Figure 3.

The new features added to the replicated structure dene the coupling structureFigure 4. Replication of the basic dynamic module to model higher maturity processes.190 RUIZ, RAMOS, AND TOROTable 1. Main differential equationsfor the basic dynamic module for

software development modeling (maturity level 12)Pending taskst = [integraltext]Initial size

0 (revision ratet development ratet) dt

Acomplished taskst = [integraltext] (development ratet revision ratet) dt

Revision pending taskst = [integraltext] (quality ratet revision ratet) dt

quality ratet = development ratet (1 QUALITY)that joinstogether the four dynamic modules. Thiscoupling structure ismainly

composed of a matrix that keeps the order of precedence among the phases and

the percentage of a task that have to be accomplished before starting the following

one. The accomplished task fraction of each phase is the value that each dynamic

module shareswith the othersaswell asthe common knowledge of the matrix of

precedence.Table 2 showsthe new form of the equationsfor thislevel. It isimportant to notice

here the appearance of the double index to identify the phase of the project in

contrast with the equations shown in Table 1, and also the fact that these equations

are only calculated when the conditionsare favourable according to the matrix of

precedence.5. DIFSPI implementationThe former conceptual ideashave been implemented to develop a tool using JavaTM

technology (Java 2 SDK v. 1.2 and 1.3). Three groupsof initial parametersare

required to drive a simulation run: parameters related to the organization (delays,

average accepted overwork, etc.), parametersrelated to the project (size, quality

objective, initial staff, etc.), and parameters to drive the simulation run. With these

initial data, it is possible to run a rst simulation to establish a baseline to the

project. The results obtained can be graphically displayed in order to merge in a

single view the static data offered by the traditional models with the dynamic data

provided by the simulation runs. After this, it is possible to experiment with different

processimprovementsand alternative plansby changing the valuesof the parameter(s) required and running the new simulations. All the results obtained are saved

in the database. This database may be used then to feed some machine learning

algorithmsin order to automatically obtain management and processimprovement

rules(Ramoset al., 2000).Table 2. Main differential equationsof the replicated module for software development modeling (maturity level 23)Pending tasksi t = [integraltext]Initial sizei

0 (revision ratei t development ratei t) dt

Acomplished tasksi t = [integraltext] (development ratei t revision ratei t) dt

Revision pending tasksi t = [integraltext] (quality ratei t revision ratei t) dt

quality ratei t = development ratei t (1 QUALITY)FRAMEWORK FOR SOFTWARE PROCESS IMPROVEMENT 191Table 3. Real and simulated data for the case studySize of the project = 80,000 LOCREAL DATA SIMULATED DATATime 250 days Time 263 days

Initial workforce 8 technician Effort 4,361 technician-day

Effort 4,780 technician-day Quality 80% (tasks revised)Workforce 9 technician5.1. DIFSPI utilizationThe potential applicationsof the DIFSPI have already been mentioned in the former sections. In this section some of the data obtained when DIFSPI was applied

inside a local software development organization are provided. This local organization could be placed at level 1. At rst the software process capability of this organization wasunpredictable because it wasconstantly changed or modied asthe work

progressed. Performance depended on both the capabilities of the project manager

and the technical team. Moreover, there were few stable software processes in evidence. According to level 1 organizations, the software process here was perceived

asan amorphousentity, a black box, and visibility into the projectsprocesses

wasvery limited. Requirementsowed into the software processin an uncontrolled

manner, giving a product asa result. The purpose of thisapplication wasto ensure

that the framework could reproduce the behavior observed in a real project and,

therefore, could trigger a metricscollection program and help in decision making,

predicting, and cost estimating. Table 3 shows the characteristics of the project that

was simulated for this case study together with the data of the baseline reported

by the simulation. It should be noted here that the data reported by the simulation

conformsthe core measuresrecommended by the Software Engineering Institute

(SEI) (Carleton et al., 1992).The scenario shown in Table 4 helps to analyze the impact of the size of the technical staff over the main four variables (time, effort, quality, and overall workforce).

Two different cases were simulated. The rst one (CASE 1) had a schedule of 250

days and 16 part-time technicians. The second case (CASE 2) had a schedule of

150 daysand 16 full-time technicians.The expected behavior for projectswith a high level of personnel isthat the average productivity per technician achieved will be lower. The average productivity

per technician in the baseline was 0.8926 tasks/(technician day). CASE 1 and 2Table 4. Simulated data for scenario analysisCASE 1 CASE 2Time 135 days Time 140 days

Effort 1,396 technician-day Effort 3,596 technician-day

Quality 91% Quality 91%Workforce 18 technician Workforce 16 technician192 RUIZ, RAMOS, AND TOROboth had double the initial workforce than that of the baseline, although schedulesand resource allocation were different between them. The average productivity

obtained for CASE 1 and 2 was, respectively, 0.8277 tasks/(technician day) and0.8142 tasks/(technician/day).6. ConclusionsMotivated by lessons learnt from another System Dynamics application in an industrial environment, the development of a framework to combine the traditional estimation toolswith the dynamic approach hasbeen initiated. The main objective of

thisdynamic framework isto assessproject managersand membersof the SEIG to

dene, evaluate, and implement processimprovementsto achieve higher levelsof

maturity. The whole process of development of the framework also helps to design

a specic metricscollection program which, once implemented, contributesto build

and feed a historical database inside an organization.With the application of DIFSPI in a level 1 organization, important benetswere

obtained. First, it must be mentioned that during the process of model building, the

project manager gained much new insight into those aspects of the development

process that mostly inuence the success of the project (time, cost, and quality).

Second, having the possibility of gaming with the DIFSPI allowed him to better

understand the underlying dynamics of the software process. As a consequence, several process improvement suggestions were easily designed and, most importantly,

analyzed using the simulation of scenarios. Finally, templates and guidelines for

a metricscollection program were almost automatically derived from the requirementsof the dynamic modules.Our future work will mainly concentrate on research towards a full development

of the dynamic modulesthat implement the key processareasof the higher maturity

levels. Once this development has been accomplished, it is intended to validate the

complete DIFSPI in real industrial environments.AcknowledgementsThe authors wish to thank the Comisin Interministerial de Ciencia y Tecnologa,

Spain, (undergrant TIC2001-1143-C03-02) for supporting this research effort.ReferencesCarleton, A., Park, R.E., Goethert, W.B., Florac, W.A., Bailey, E.K., and Peeger, S.L. 1992. Software measurement for DoD systems: Recommendations for initial core measures, Technical Report

CMU/SEI-92-TR-19. Software Engineering Institute, Pennsylvania, Pittsburgh, Carnegie Mellon University.Christie, A.M. 1999. SimulationAn enabling technology in software engineering. http://www.sei.cmu.

edu/publications/articles/christie-apr1999/christie-apr1999.html.Christie, A.M. 1999. Simulation in support of CMM-based process improvement, The J. of Systems andSoftware, 46(2/3): 107112.Eberlein, R.L. 1989. Simplication and understanding of models, Systems Dynamics Review. 1(5): 5168.FRAMEWORK FOR SOFTWARE PROCESS IMPROVEMENT 193High Performance Systems, Inc., 2001. ithink 7.0, New Hampshire, Hanover.

Kellner, M.I., Madachy, R., and Raffo, D. 1999. Software process simulation modeling: Why? What?How? The J. of Systems and Software. 46(2/3): 91105.Paulk, M., Garcia, S.M., Chrissis, M.B., and Bush, M. 1993. Key practices of the capability maturity

model, Version 1.1, Technical Report CMU/SEI-93-TR-25. Software Engineering Institute, Pennsylvania, Pittsburgh, Carnegie Mellon University.PowerSim Corporation, 2001. Powersim Studio 2001. Virginia, Hendon.

Putnam, L.H. 1992. Measures for Excellence: Reliable Software, On Time, within Budget, New York,Prentice-Hall.Ramos, I., Aguilar, J., Riquelme, J.C., and Toro, M. 2000. A new method for obtaining software project

management rules, SQM 2000, United Kingdom, Greenwich, pp. 149160.Rodrigues, A. and Bowers, B. 1996. System dynamics in project management: A comparative analysis

with traditional methods, System Dynamics Review. 12(2): 121139.Ruiz, M., Ramos, I., and Toro, M. 2001. A simplied model of software project dynamics, The J. ofSystems and Software. 59: 299309.Ventana Systems, Inc., 2002. Vensim Version 5. Massachusetts, Harvard.Mercedes Ruiz received the B.Sc. degree in Computer Science from the University of Seville, Seville,

Spain, in 1995. She is currently working towards her Ph.D. at this University. She is an assistant Professor

of Information Systems, Department of Computer Languages and Systems, University of Cdiz. Her

works have appeared in several refereed journals and she is the author of several papers, books and

chapters of books. Her research interests are in software process simulation models, software metrics

and quality, and decision support for dynamic tasks.IsabelRamos wasborn in Seville in 1957. She hasa Ph.D. in Industrial Engineering from the University

of Seville where she works teaching Software Engineering, in the Department of Computer Languages

and Systems, University of Seville. She has specialized in software development projects management,

and currently isthe manager of FIDETIA (Foundation for Research and Development on Information

Technology in Andaluca).194 RUIZ, RAMOS, AND TOROMiguelToro obtained an Industrial Engineerstitle and DoctorsIndustrial Engineer title from the

University of Seville. At the present time, he is a Professor in the Department of Computer Languages

and Systems, University of Seville, and is the manager of the Ofce of Transfer of the Investigation

Results of the University of Seville (OTRI). His research activity is mainly in two lines: Qualitative

Reasoning and Learning, and Software Engineering. He is author or joint author of a large amount of

works, including articles of magazines (national and international), congresses, books and chapters of

books. He ismember of ACM (groupsof interest SIGSOFT, SIGART), IEEE Computer Society and

other scientic societies.