Implementing cloud-based parallel metaheuristics: an overview

Abstract

Metaheuristics are among the most popular methods for solving hard global optimization problems in many areas of science and engineering. Their parallel implementation applying HPC techniques is a common approach for efficiently using available resources to reduce the time needed to get a good enough solution to hard-to-solve problems. Paradigms like MPI or OMP are the usual choice when executing them in clusters or supercomputers. Moreover, the pervasive presence of cloud computing and the emergence of programming models like MapReduce or Spark have given rise to an increasing interest in porting HPC workloads to the cloud, as is the case with parallel metaheuristics. In this paper we give an overview of our experience with different alternatives for porting parallel metaheuristics to the cloud, providing some useful insights to the interested reader that we have acquired through extensive experimentation.