Solving Graph Partitioning Problems with Parallel Metaheuristics

In this article we describe computer experiments while testing a family of parallel and hybrid metaheuristics against a small set of graph partitioning problems like clustering, partitioning into cliques and coloring. In all cases the search space is composed of vertex partitions satisfying specific problem requirements. The solver application contains two sequential and nine parallel/hybrid algorithms developed on the basis of SA and TS metaheuristics. A number of tests are reported and conclusions concerning metaheuristics' performance that result from the conducted experiments are derived. The article provides a case study in which partitioning numbers psi(k)(G), k >= 2, of DIMACS graph coloring instances are evaluated experimentally by means of H-SP metaheuristic which is found to be the most efficient in terms of solution quality.