Scheduling a job mix in a partitionable parallel system

Efficient scheduling of jobs on parallel processors is essential for good performance. However, design of such schedulers is challenging because of the complex interaction between system and workload parameters. This paper studies the performance of a partitionable parallel system in which job scheduling depends on job characteristics. Jobs consist of a different number of tasks and are characterized as sequential or parallel depending on whether the tasks are processed sequentially on the same processor or at different processors. Jobs that consist of parallel tasks are called gangs, that is, they have to be scheduled to execute concurrently on processor partitions, where each task starts at the same time and computes at the same pace. The goal is to achieve good performance of sequential and parallel jobs. The performance of different scheduling schemes is compared over various workloads. Simulated results indicate that sequential jobs should not arbitrary overtake the execution of parallel jobs.