Token space minimization by simulated annealing

We describe a heuristic solution for the minimum token space scheduling (MTSS) problem, based on simulated annealing. In MTSS, one schedules a set of tasks with precedence constraints, represented by a directed graph. The arcs in the graph represent data, or tokens, which the tasks must receive before they can be processed. MTSS seeks to minimize the maximum number of tokens extant at any time during execution, while minimizing completion time. We motivate MTSS with an application from computer architecture: maximizing the locality of data required for execution of a program by multiprocessors. Simulation results demonstrating the effectiveness of our method are presented.