Parallel Heuristic Search - New Approaches to Parallel Simulated Annealing and Tabu Search

A large number of combinatorial optimization problems can presently only be solved to proven optimality if they are of very moderate size. This fact motivates the interest in approximation algorithms and heuristics. The former class of algorithms are able to guarantee the quality of the produced (suboptimal) solutions within certain bounds, while heuristics usually only have empirical evidence of their problem solving abilities. Many of these heuristics are tailored to a specific problem type. There exists however a small family of more generally applicable heuristics (metaheuristics), which with a little effort can be turned into a heuristic for a variety of problems. Belonging to this family are Simulated Annealing [3] and Tabu Search [3]. Even though heuristics offer the ability to find good solutions in moderate time, the computational expense may still be significant. It is therefore relevant to try to exploit the potentially vast computational power of parallel computers by designing and implementing parallel versions of the heuristics. Parallelization of Simulated Annealing has received some attention so far [2], while parallelization of Tabu Search has only recently received noteworthy attention [1].