Pipeline for parallel implementation of multi-objective particle swarm optimization

Communication overhead is an issue in a parallel implementation of algorithm on loosely coupled systems. Pipeline is a technique used to achieve parallelism at instruction level by dividing task into different stages and for every stage, it´s output is the input for next the stage. Thus, pipeline limits the communication overhead by restricting it to only for previous and next stage. Pipeline also performs faster by overlapping the instructions. In this paper, we propose new approach for parallel implementation of multi-objective particle swarm optimization (MOPSO) using logical pipeline, which focuses on overlapping the execution of functions. Logical pipeline for MOPSO is the disposition of three stages, each for calculating velocity, evaluation and updating of archive. We compared the results with the different parallel models for MOPSO and improvement in time is significant.