Publication

Abstract : A grid system aggregates a collection of heterogeneous resources throughout the globe to perform complicated computations in large scale. In this paper, an improved scheduling methodology is proposed with three different scheduling policies. The first scheduling strategy schedules the jobs to the resources by dividing the jobs in to multiple chunks and assigning iteratively to all the clusters. The second strategy uses a resource broker as a middleware to route jobs randomly to all its computing elements. The third strategy proposes an advanced routing policy to route jobs to the clusters which comprises of log entries computed at each cluster and the same is updated in the resource broker so that the resource broker can further route jobs to clusters which are identically found with lesser load by comparing the status of the log entries. This log also helps to identify the faulty grid resources and hence to make the system more effective a fault-log is also maintained in the Grid Information Server (GIS). The behaviour of grid system is analyzed by considering several metrics for a huge variety of workload configurations such as work load, distribution of jobs, number of clusters, cluster size etc. With a typical grid simulation, the performance of the grid with the proposed system is compared with that of most of the classical brokering strategies. The proposed system claims a significant and consistent improvement in its performance in terms of execution time of jobs with varying loads which inturn helps the system to achieve optimum load balancing and fault tolerance with respect to their corresponding log entries. © Research India Publications.