A Simulation Study of Epoch Load Sharing in Distributed Systems

Abstract

This paper examines load sharing in distributed systems. It proposes a special load-sharing method, referred to as epoch load sharing. With this policy, load is evenly distributed among distributed processors, with job migration only at the end of predefined intervals. The time interval between successive load sharing is called an epoch. The authors compare the performance of epoch job migration with other traditional load-sharing methods. The objective is to reduce the number of times that global system information is needed to make allocation decisions while achieving good overall performance and fairness in individual job service. The impact of the variability in migration cost and job service demand is also examined. Simulation results indicate that for specific workloads, epoch job migration can combine good performance with a moderate scheduling overhead.