Premptive Job Scheduling with Priorities and Starvation cum Congestion Avoidance in Clusters

Abstract

This paper describes a new policy to schedule parallel jobs on Clusters that may be part of a Computational Grid. This algorithm proposed 3 Job Queues. In each Cluster, some number of resources is assigned to each of the Queue. The 1st Queue has some jobs which has low expected execution time(EET). The 2nd Queue has some jobs which has high expected execution time. The 3rd Queue has jobs which are part of Meta-Job from Computational Grid. In 1st there is no chance of starvation. But in 2nd Queue there is a chance of starvation. So this algorithm applied Aging technique to preempt the jobs which has low priority. And the 3rd Queue is fully dedicated to execute a part of Meta-Jobs only. So here we maintain multiple job Queues which are effectively separate jobs according to their projected execution time for Local Jobs and for part of Meta-Job. Here we preempt jobs by applying Aging Technique. Here we can avoid unnecessary traffic congestion in networks by comparing Expected Execution Time with Total Time for submitting job(s) and receiving result(s) from node(s).