A Globus-Based Distributed Enumerative Search Algorithm for Multi-objective Optimization
Francisco Luna, Antonio Jesús Nebro, Enrique Alba · 2004
Abstract — Enumerative search is a technique to solve multi-objective optimization problems based on evaluating each pos-sible solution from a given finite set. The technique is simple and computationally expensive, but it is the only way at present to compute exact Pareto fronts in multi-objective problems. In this context, Grid computing systems offer a potentially large amount of computing power that can be used to overcome the mentioned drawback to some extent. In this paper, we analyze several practical and technical issues concerning the use of the Globus Toolkit, a de facto standard system for Grid computing, to implement a distributed enumerative search algorithm. We have solved a benchmark of multi-objective problems in a cluster of computers and we have analyzed issues such as the parallel efficiency, mean CPU use, and network bandwidth utilization. Furthermore, we also use Globus to develop a new technique named grid-µGA, an extension of the micro-GA algorithm. The results indicate that using Globus is a promising choice to solve multi-objective problems in Grid computing systems. I.