Model-Based Scheduling Analysis for Software Projects

We show how to compute optimal policies for the scheduling of software development projects under uncertainty. Our approach is based on a stochastic scheduling model that explicitly captures the strong feedback between the tasks in software development ("ripple effects"). We apply reinforcement learning to the optimization problem. For a selected sample project, we compute the optimal policy, simulate the project, and analyze the task assignments that are made by the optimal policy. From the analysis of the simulated schedules, we derive tentative, generic scheduling guidelines. The guidelines prioritize certain tasks based on the characteristics of the software architecture, and assign tasks according to the past performance of the developers.