Gang Scheduling Performance under Different Distributions of Gang Size

Abstract

In this paper we study gang scheduling in a partitionable parallel system. System performance is examined and compared in cases where different distributions for the number of parallel tasks per job (gang size) are employed. We examine cases where gang size is defined by a specific distribution and also a case where gang size distribution varies with time. A simulation model is used to address performance issues associated with gang scheduling. Four scheduling policies are employed to schedule parallel jobs for a variety of workloads. Simulated results reveal that the scheduling policies exhibit similar behavior in all cases of distributions of gang size that we examined.