Charles B. McVey, Ella M. Atkins, Edmund H. Durfee, Kang G. Shin
The traditional scheduling problem can be reformulated in an agent-oriented paradigm. A manager agent proposes a set of tasks to schedule along with scheduling constraints among these tasks. A scheduler agent receives this data and attempts to build a schedule which satisfies all scheduling constraints. Ideally, if the scheduler solves the problem as the manager requests, a successful schedule will be returned. If the scheduler is unable to satisfy the manager’s request, it might return NULL, forcing the manager to blindly modify and resubmit its request. Our scheduler is designed to return feedback to the manager regarding why a successful schedule could not be produced from the original request. This additional knowledge can then be used by the manager to direct its modification of tasks and/or relaxation of constraints for a subsequent scheduling request. This process iterates until a successful schedule with well-utilized resources is obtained.