Agent-mediated genetic super-scheduling in grid environments

Abstract

Super-scheduling in a dynamic grid environment is a very challenging issue that remains to be solved before a grid can be deployed and effectively utilized. In this paper we investigate a paradigm based on genetic algorithms (GA) to efficiently solve the scheduling problem. This GA paradigm is architecturally combined with the multiagent system (MAS) paradigm to form a flexible super-scheduling system. A three-layered scheduling architecture is presented and the corresponding realization of a multiagent-based system is described. The experiment shows that the better scheduling results are obtained for the adopted metrics of flow time and job stretch.