Optimal scheduling of aperiodic jobs on cluster

Abstract

This paper presents an algorithm to schedule aperiodic jobs on a cluster system. A cluster consists of one dispatcher and multiple schedulers. The algorithm is optimal for the response time of each aperiodic job. Aperiodic jobs are partitioned into phases and assigned deadlines by the dispatcher. With the deadline, the aperiodic jobs can be scheduled uniformly under EDF (Earliest Deadline First) together with periodic jobs. With rigorous theoretical analysis and proof, we have that even if the dispatcher has no scheduling information, the algorithm can still dispatch the aperiodic jobs onto the proper scheduler to achieve optimal response time and also guarantee the deadline requirements of periodic jobs.