On multiprocessor scheduling with cellular automata

In this paper we propose using cellular automata (CAs) to perform distributed scheduling tasks of a parallel program in a multiprocessor system. We consider a program graph as a CA with elementary cells interacting locally according to a certain rule which must be found. Effective rules for a CA are discovered by a genetic algorithm (GA). With these rules, CA-based scheduler is able to find allocations which minimize the total execution time of the parallel program in the two processor system. We analyse different modes of operating of the scheduler in the case of two processors. We also show how our algorithm works in the case of more than two processors.