Particle Swarm Optimization for the Exploration of Distributed Dynamic Load Balancing Algorithms

Abstract

Evolutionary algorithms provide mechanisms that can achieve efficient exploration for complex design spaces. Also, they constitute an efficient tool for identifying the best alternatives to implement the solution of a certain problem. In this work we use particle swarm optimization (PSO) to find the best alternatives for the distributed load balancing procedure in heterogeneous parallel computers. We have classified and parameterized the different distributed strategies of the dynamic load balancing, then we have applied a methodology based on PSO capable of analyzing the characteristics of the alternatives of load balancing when considering different types of problems and parallel platforms. As an application example of the proposed methodology we will show the results corresponding to the dynamic load balancing in a heterogeneous cluster of PCs for a parallel branch and bound algorithm