Effectiveness of Heuristics and Simulated Annealing for the Scheduling of Concurrent Tasks - An Empirical Comparison

It is well-known that the scheduling of concurrent tasks with precedence constraints in parallel systems in order to minimize the makespan is NP-complete. We study both the average effectiveness and the average efficiency of 27 heuristics and 7 simulated annealing algorithms used for the minimization of makespan. It is shown, by a computational experiment, that the simulated annealing algorithms are very effective compared with the heuristics, provided these algorithms converge. It turns out that some heuristics are quite effective on the average, and that the heuristics, provided they are used together, have a qualitative behavior not much worse than that of the simulated annealing and are much more efficient.