Genetic algorithm calibration for two objective scheduling parallel jobs on hierarchical grids

Abstract

This paper addresses non-preemptive offline scheduling parallel jobs on a Grid. We consider a Grid scheduling model with two stages. At the first stage, jobs are allocated to a suitable Grid site, while at the second stage, local scheduling is independently applied to each site. In this environment, one of the big challenges is to provide a job allocation that allows more efficient use of resources and user satisfaction. In general, the criteria that help achieve these goals are often in conflict. To solve this problem, two-objective genetic algorithm is proposed. We conduct comparative analysis of five crossover and three mutation operators, and determine most influential parameters and operators. To this end multi factorial analysis of variance is applied. © 2012 Springer-Verlag.