Addressing sporadic contention on shared computing clusters

In a virtual organization, physical organizations share their resources when organizations to save on their investment of computing resources. Sporadic contention is caused by on-demand requests when organizations seek to consume large amounts of resource from each other. For this reason, admission control mechanisms need to be incorporated to prevent oversubscription of shared resources. The major constrain in admission control is that requests are handled FCFS. As such, the service-level agreement (SLA) cannot be changed by the system at the discretion of the resource manager. We characterize this problem as non-preemptive admission control. In this paper, we introduce a resource scheduling framework for admission control in a grid computing environment for this purpose. It is built on a reference pricing model and utility functions. To evaluate the admission control scheme, we compared its scheduling performance with known deadline admission control algorithms