Scheduling Jobs with Estimation Errors for Multi-server Systems

Abstract

When scheduling single server systems, Shortest Remaining Processing Time (SRPT) minimizes the number of jobs in the system at every point in time. However, a major limitation of SRPT is that it requires job processing times a priori. In practice, it is likely that only estimates of job processing times are available. This paper proposes a policy that schedules jobs with estimated job processing times. The proposed Modified Comparison Splitting Scheduling (MCSS) policy is compared to SRPT when scheduling both single and multi-server systems. In the single server system we observe from simulations that the proposed scheduling policy provides robustness that is crucial for achieving good performance. In contrast, in a multi-server system we observe that robustness to estimation errors is not dependent on the scheduling policy. However, as the number of servers grows, SRPT becomes preferable.