On-line evolutionary resource matching for job scheduling in heterogeneous grid environments

Abstract

In this paper, we describe a resource matcher (RM) developed for the on-line resource matching in heterogeneous grid environments. RM is based on the principles of Evolutionary Algorithms (EA) and supports dynamic resource sharing, job priorities and preferences, job dependencies on multiple resource types, and resource specific and sitewide policies. We describe the evolutionary algorithm and the models used for representing the resource requirements, preferences, and policies. We evaluate three different methods for bootstrapping RM. We then describe a Evolutionary Matcher (EM) Service - a Web Service implementation of RM for on-line resource matching. Preliminary performance results indicate that the EM service is efficient in speed and accuracy and can keep up with high job arrival rates - an important criterion for on-line resource matching systems. The service oriented architecture makes the EM service scalable and extensible and can be integrated with already existing grid services in a straightforward manner. © 2006 IEEE.