Hybrid MPI/openmp parallel evolutionary algorithms for vehicle routing problems

Abstract

The traditional fields of improvement in parallelism have been orientated to experimentation on high-budget equipment, such as clusters of computers or shared memory machines thanks to their highperformance and scalability. In recent years, the generalization of multicore microprocessors in almost all the computing platforms makes it possible to take advantage of parallel processing even for the desktop computer user. This paper analyzes how to improve the performance of population-based meta-heuristics using MPI, OpenMP, and hybrid MPI/OpenMP implementations in a workstation having a multi-core processor. The results obtained when solving large scale instances of the Capacitated Vehicle Routing Problem with hard Time Windows (VRPTW) show that, in all cases, the parallel implementations produce better quality solutions for a given amount of runtime than the sequential algorithm, and also solutions of similar quality in less runtime.