Task scheduling on minimal processors with genetic algorithms

This paper addresses the problem of scheduling parallel program represented as directed acyclic task graphs onto multiprocessors. The task scheduling problem is known to be NP-complete. Existing task scheduling algorithms either are assumed on a bounded number of processors, or generate schedules that need more processors than necessary. Genetic algorithms are successfully applied to solve the problem of scheduling parallel program tasks on a fixed number of processors. Task duplication is also an effective technique for shortening parallel execution time of program. Meanwhile, this technique generates useless task duplications. In this paper, we propose a GA based algorithm to solve the task scheduling problem. Our algorithm can yield a schedule with shorter parallel execution time and fewer required processors, and without useless task duplications. We compare our algorithm with GA based algorithm. Experimental results show that our algorithm outperforms it when communication delay is large.