Decentralized dynamic load balancing for multi cluster grid environment

Abstract

Load balancing is essential for efficient utilization of resources and enhancing the performance of computational grid. Job migration is an effective way to dynamically balance the load among multiple clusters in the grid environment. Due to limited capacity of single cluster, it is necessary to share the underutilized resources of other clusters. Each cluster saves the static and dynamic information about its neighbors including transfer delay and load. This paper addresses the issues in multi cluster load balancing based on job migration across separate clusters. A decentralized grid model, as a collection of clusters for computational grid environment is proposed. A Sender Initiated Decentralized Dynamic Load Balancing (SI-DDLB) algorithm is introduced. The algorithm estimates system parameters such as resource processing rate and load on each resource. The algorithm balances the load by migrating jobs to the least loaded neighboring resource by taking into account of transfer delay. The algorithm also considers the availability of selected resource before dispatching job for execution since the probability of failure is more in the dynamic grid environment. The main goal of the proposed algorithm is to reduce the response time of the jobs. The proposed algorithm has been verified through the GridSim simulation toolkit. Simulation results show that the proposed algorithm is feasible and improves the system performance considerably. © Springer-Verlag Berlin Heidelberg 2011.