The main issue of parallel job scheduling is how to share the resources of the parallel machine among a number of competing jobs. The most commonly used scheduling scheme is variable partitioning that allocates each job a partition of the machine for its use. But this approach suffers from fragmentation because partitions are allocated in a first-come first-served(FCFS) manner.
Backfill scheduling address this problem by allowing jobs to move ahead in the queue only if it does not delay the higher-priority jobs. The performance of backfill scheduling scheme depends on the job characteristics and priority. Until now, there are many studies for improving performance of backfill scheduling. But most research has focused on either system utilization or job slowdown.
We adjust priority of jobs with consideration of trade-offs between system utilization and job slowdown. First, we propose group-based backfill scheduling that categorize jobs on their processor request size and job execution time, and prioritize them according to categorized result. Next we suggest the adaptive reservation scheme that allows jobs to move ahead if it does not delay the selected set of jobs to reduce the job slowdown.
Simulation results show that proposed scheme reduces total job execution time by improving system utilization and prevents job from increasing slowdown.
