A new heuristic algorithm based on GAs for multiprocessor scheduling with task duplication

In this paper, we propose a new algorithm for scheduling parallel programs represented as directed acyclic graphs onto multiprocessors with communication delays. In such systems, task duplication is known as a useful technique for shortening the length of schedules. The proposed algorithm adopts several heuristics based on GAs as well as task duplication. To apply a GA to scheduling, we design chromosomes using list representation so that each chromosome can uniquely represent a schedule of tasks. We also design genetic operators to control the degree of replication of tasks. Through simulation studies for three kinds of parallel programs under various scheduling conditions, we compare the proposed algorithm with an established algorithm proposed by Kruatrachue. As a result, it is found that the new heuristic algorithm outperforms the previous algorithm especially when communication delays are relatively small.