Heterogeneous Parallel Implementation of PSO Algorithm with an Aging Leader and Challengers

Abstract

Parallel computing techniques provide high-performance execution of large-size and real-world complicated problems such as heuristic optimization algorithms. One of the most popular optimization algorithms is PSO (particle swarm optimization) which is an intelligent search method for finding the best solution according to population state. The PSO algorithm has obtained significant attention from researchers and different parallel implementations of this algorithm have been presented for intensive-computing applications. The ALC-PSO algorithm (PSO with an aging leader and challengers) is an improved population-based and intensive-computation procedure thanks to the high fitness analyses. Compared to the traditional PSO, the ALC-PSO algorithm improves convergence rapidity. In this paper, we have developed a novel heterogeneous parallel implementation of the ALC-PSO algorithm using OmpSs and CUDA, for execution on both CPU and GPU cores. OmpSs is a task-based parallel programming model that uses runtime system and hardware for automatic exploiting and managing parallelisms. Combination of OmpSs with CUDA helps developers accelerate their applications running on both CPUs and GPUs. The results demonstrate that the proposed approach (i.e., the OmpSs-CUDA implementation of the ALC-PSO algorithm) provides higher performance than the serial, and also the CUDA-based parallel implementations of the ALC-PSO algorithm.