Load Balancing in Clustersusing optimal Resource Set of Nodes

A probability model for job allocation to nodes in a cluster for better CPU utilization and reduces I/O and communication overhead is presented in this paper. For the NP-hard problem of job assignment, load balancing is done with a local and global scheduler (LS and GS). A newly arriving job need not wait at the home node for the want of resources as GS takes care of it. Using three different queues for the processes, GS assigns the best set of optimal resources at a foreign node with a minimal transfer cost to the processes. The experimentally simulated results show that the model compares well with the existing models and well suited for time and space sharing systems with minimal I/O overhead and excellent resource utilization.