Optimising parallel applications on the Grid using irregular array distributions

Abstract

In this paper we propose a new approach for scheduling data parallel applications on the Grid using irregular array distributions. We implement the scheduler as a new case study for using a general purpose experiment management tool that we developed in previous work for performance tuning and optimisation of scientific applications. We report results on scheduling a Jacobi relaxation application on a simulation testbed of the Austrian Grid [2] using a problem independent plug-and-play genetic algorithm. © Springer-Verlag Berlin Heidelberg 2005.