Scheduling Gangs in a Distributed System

In this paper we study the performance of parallel job scheduling in a distributed system. A special type of scheduling called gang scheduling is considered. In gang scheduling jobs consist of a number of interacting tasks, which are scheduled to run simultaneously on distinct processors. Two gang scheduling policies are used to schedule parallel jobs for two different types of job parallelism. Also, we present the average performance of all jobs as well as the relative performance of small and large gangs. We examine various workloads using simulation techniques. The results show that the policy, which gives priority to large gangs performs better than the one that does not take job characteristics into account while making the ordering decision.