Abstract. We present a parallel job scheduling approach for coarsegrain timesharing which preempts jobs to disk and avoids any additional memory pressure. The approach provides control regarding the resource shares allocated to different job classes. We demonstrate that this approach significantly improves response times for short and medium jobs and that it permits controlling different desirable resource-shares for different job classes at different times of the day according to site policies.
Bryan Esbaugh, Angela C. Sodan