Abstract
We examine the optimal allocation of engineering resources in software maintenance operations. We first present competing theoretical models of software defect resolution. Next, we estimate these models using field data on defect resolution for 12503 defects that occurred over 97 months in a large systems software product. We then apply the most appropriate model - one based on a beta-geometric distribution - to estimate the level of engineering resources
required to manage debugging efforts. We propose heuristics based on temporal cut off policies to maximize the rate of successful resolution of bugs. We empirically demonstrate that such policies can be applied to enhance productivity in this complex environment. In particular, we show that the proposed
cut off policies minimally impact rates of successful resolution while substantially reducing waiting times for incoming bugs in the system, thereby improving overall productivity and resource utilization. Finally, we extend our model to consider two alternative scenarios where (a) incoming bugs differ on
priority levels assigned for resolution and (b) incoming bugs are from different populations.
This is joint work with Sriram Narayanan at Michigan State University and Sridhar Balasubramanian at University of North Carolina at Chapel Hill.