A grid scheduling based on generalized extremal optimization for parallel job model

Abstract

In this paper we propose an efficient parallel job scheduling algorithm for a grid environment. The algorithm implies two stage scheduling. At the first stage, the algorithm allocates jobs to the suitable machines, where at the second stage jobs are independently scheduled on each machine. Allocation of jobs on the first stage of the algorithm is performed with use of a relatively new evolutionary algorithm called Generalized Extremal Optimization (GEO). Scheduling on the second stage is performed by some proposed heuristic. We compare GEO-based scheduling algorithm applied on the first stage with Genetic Algorithm (GA)-based scheduling algorithm. Experimental results show that the GEO, despite of its simplicity, outperforms the GA algorithm in all range of scheduling instances. © 2012 Springer-Verlag.