Stochastic analysis of scheduling strategies in a Grid-based resource model

Abstract

A model inspired by a scenario found in Grid-based scheduling systems is considered. Scheduling is performed remotely without access to up-to-date resource availability and usage information. This system is modelled as a collection of queues where servers break down and are subsequently repaired. There is a delay before the scheduler learns of failures, and requests may continue to arrive into a resource queue for some time after active service has ceased. The queues are considered to be persistent under failure. However, these queues have finite capacity; therefore there is the possibility that queues become full, causing job-loss. Stochastic process algebra and stochastic probes are used to analyse this model to find steady-state measures and passage time distributions. The effect of the duration of any delay on information propagation on the system response time and job loss is investigated and evaluated numerically.