Scheduling strategies for multitasking in a distributed system

Abstract

We study various policies for scheduling tasks in distributed processor queues. Their performance is studied and compared for a variety of workloads. It is our intention to find a policy that increases throughput and also is fair to jobs. Simulation results indicate that the policy that schedules the shortest task in a queue, when there is not any other task that has been waiting more than some configurable period of time, yields good system performance and also provides a guarantee for fairness in individual job execution.