A High-Performance Parallel Implementation of ALC-PSO Algorithm using OmpSs and CUDA

Parallel computing techniques provide high-performance execution of the large-size real-world complex problems such as heuristic optimization algorithms (e.g., PSO). PSO, as a population-based stochastic search technique for solving optimization problems, has been proven to be effective in a wide range of applications. Hence, parallel implementation of this algorithm with numerous parallelization models and strategies for solving complex applications has obtained significant attention by researchers. In comparison with the PSO, ALC-PSO algorithm avoids the problem of premature convergence and improves convergence speed. Similar to most of the evolutionary algorithms, ALC-PSO is population-based iterative and computationally intensive, because the optimizing process of this algorithm requires a large number of fitness evaluations which can run sequentially on CPU or parallel on GPU or on both of them. In this article, we propose parallel implementation of ALC-PSO algorithm using OmpSs and CUDA, two powerful parallel programming models, for parallel executing the algorithm on both CPUs and GPUs. OmpSs is a task-parallel programming model and helps developers accelerate their applications execution, on CPU and GPUs, with a low programming effort. The results show the proposed implementation provides higher performance than the serial and the CUDA-based parallel implementations of ALC-PSO.

Keywords:

Parallel programming , ALC-PSO , GPU , OmpSs , CUDA , Performance.