Periodic scheduling of mixed workload in distributed systems

Abstract

Efficient scheduling of jobs in distributed systems is essential to achieving good performance. This paper addresses issues of scheduling a job mix, consisting of parallel jobs and single-task, but computationally intensive jobs, on distributed processors. It considers a scheduling algorithm, referred to as periodic scheduling. With this algorithm, processor queues are rearranged only at the end of predefined time intervals, called periods. The goal is to examine whether periodic scheduling can significantly reduce the overhead that is involved with the corresponding non-periodic policy and to also provide good performance. Simulation is employed to evaluate the performance of this algorithm for different period sizes. Simulation results demonstrate that periodic job scheduling can meet the goal.