Superlinear Speedup Of Parallel Population-Based Metaheuristics: A Microservices And Container Virtualization Approach

Population-based metaheuristics such as Evolutionary Algorithms (EAs) can require massive computational power for solving complex and large scale optimization problems. Hence, the parallel execution of EAs attracted the attention of researchers as a feasible solution in order to reduce the computation time. Several distributed frameworks and approaches utilizing different hardware and software technologies have been introduced in the literatures. Among them, the parallelization of EAs in cluster and cloud environments exploiting modern parallel computing techniques seems to be a promising approach. In the present paper, the parallel performance in terms of speedup using microservices, container virtualization and the publish/subscribe messaging paradigm to parallelize EAs based on the Coarse-Grained Model (so-called Island Model) is introduced. Four different communication topologies with scalable number of islands ranges between 1 and 120 are analyzed in order to show that a partial linear/superlinear speedup is achievable for the proposed approach.