Managing risk of inaccurate runtime estimates for deadline constrained job admission control in clusters

Abstract

The advent of service-oriented Grid computing has resulted in the need for Grid resources such as clusters to enforce user-specific service needs and expectations. Service Level Agreements (SLAs) define conditions which a cluster needs to fulfill for various jobs. An example of SLA requirement is the deadline by which a job has to be completed. In addition, these clusters implement job admission control so that overall service performance does not deteriorate due to accepting exceeding amount of jobs. However, their effectiveness is highly dependent on accurate runtime estimates of jobs. This paper thus examines the adverse impact of inaccurate runtime estimates for deadline constrained job ..