SoD: Early Predictive Design Evaluation for Software

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

ABSTRACT 0438929 Mary Shaw, Ashish Arora, Shawn Butler CMU Early Predictive Design Evaluation for Software This research seeks to establish a scientific basis for analyzing designs of software-intensive systems when the code implementing the designs is unavailable. This is the situation early in system design, when decisions that profoundly affect the implementation must be made; these decisions are much easier and cheaper to change during early design than after implementation has yielded running code. By substantially improving the ability of software designers to predict properties of an implementation without actually inspecting the code, this work enables designers to understand better the consequences of early design decisions and facilitates comparison of design alternatives to a degree not currently possible. The common thread of the analysis techniques developed through this research is prediction of properties an implementation, to determine the value of the design to a client with a particular set of preferences. The techniques draw on models from economics to account for the uncertainties associated with prediction and future user needs. Predictive analysis techniques for two specific tasks are being developed: dynamic reconfiguration of mobile device application suites to satisfy changing user needs and defect profile prediction for future releases of multi-release systems. Further, the research explores a unified framework for predictive evaluations of this type