Ensemble Scheduling: Resource Co-Allocation on the Computational Grid

Abstract

This paper investigates a novel scheduling paradigm, ensemble scheduling, that can be used to schedule mixed workloads consisting of single-and multi-resource applications in Grid environments. Ensemble scheduling exploits the pattern of prior resource requests for ensemble (multi-resource) applications to dynamically partition the workload across the system resources. We show via simulation that an ensemble-aware scheduler can provide better performance both for ensemble applications and the more common single resource applications in Grid environment.