A fair and effective scheduling strategy for workstation clusters

In recent years workstation clusters have been increasingly used as general purpose computing servers for the execution of parallel and sequential applications submitted by many competing users. To make clusters a real alternative to more traditional general purpose computing platforms, scheduling techniques able to efficiently and fairly schedule collections of parallel and sequential applications must be devised. In this paper we propose a scheduling technique able to achieve the above goals by combining stride scheduling with a ticket redistribution policy that results in the spontaneous coscheduling of parallel applications. A simulation analysis carried out for a variety of workloads and operational conditions shows that our strategy outperforms previous strategies both in terms of efficiency and fairness.