Control-theoretic decision support for mitigation of modeled software project cost overruns

Scott D Miller, Purdue University

Abstract

Despite sixty years of practice, the production of software remains an endeavor that is difficult to manage according to a schedule. Control theory studies the ability to influence the dynamical behavior of systems to achieve desired behaviors or eliminate undesired behaviors. In this work, the management problem of software project schedule adherence is re-cast as a problem in control theory. Below, a modeling framework is proposed for capturing the constraints and dependencies found in the arbitrary organization specific work-flows underlying software production. Combined with proposed models for productivity, defect introduction, and defect detection, the framework completes a method for producing models of progress in software development using the techniques dynamical systems modeling. Management objectives are then formalized in terms of behaviors to be elicited or eliminated from the models. Finally, the techniques of control theory are applied to determine changes of the variable inputs of the model that maximize achievement of the management objectives. Simulation results are analyzed and reveal that the control technique succeeds in modifying the behavior of such models to improve adherence to specified management objectives. The control loop is intended to be closed by the software project manager, who translates the suggested modifications of the model inputs into candidate changes to be optionally applied to the real world software development process. As the candidate process changes enter into the decision-making process of the project manager, this work constitutes a management decision support tool.

Degree

Ph.D.

Advisors

DeCarlo, Purdue University.

Subject Area

Computer science

COinS