Improving small job response time for opportunistic scheduling

Opportunistic scheduling utilizes idle workstations to execute batch jobs. Other than ensuring that each batch user receives a fair allocation of service, no work addressing opportunistic scheduling of batch jobs has been done. The authors propose two simple scheduling policies to be used in conjunction with the Condor Up-Down scheduling algorithm. We show that an order of magnitude reduction in mean job slowdown can be achieved by rotating a user's batch jobs within their queue. Furthermore, this improved slowdown can be achieved without a significant loss of throughput. The reduced slowdowns result in a more interactive nature of the system, thus increasing its appeal. Our proposed modifications to the algorithms would require very little effort to implement.