A Probabilistic Approach for Fully Decentralized Resource Management for Grid Systems

The specific problem that underlies in collaborating Grids is scheduling of resources with no knowledge about availability of the resources due to the distributed and autonomous nature of the underlying Grid systems. In this paper, we propose a fully decentralized and probabilistic resource management scheme for Grid systems collaborating based on peer-to-peer communication paradigm. The key idea we employ is to use benchmarked performance measured about the static resource information and calculate the job execution workload. Then this benchmarked job execution time is used to predict the job scheduling feasibility in the face of resource dynamism on the target system. We design our scheme as self adjusting to the actual resource behavior and performance. Simulation results validate the appropriateness of: our scheme.