Efficiency of parallel metaheuristics for solving combinatorial problems

Abstract

The paper investigates the speedup and quality of solution of parallel metaheuristics on multicomputer platform for the case studies of parallel genetic computation for solving the TSP and solving the room assignment problem by parallel simulated annealing. Parallel computational models have been suggested for solving the TSP by genetic approach with chromosome migration (SPMD paradigm) and for solving the room assignment problem by simulated annealing (manager/workers paradigm). The experimental study is based on flat (MPI-based) parallel program implementations on multicomputer platform. Performance and scalability analysis have been made in respect to the application size and multicomputer size. The impact of various factors on the quality of solutions have been investigated and presented.