McMaster Grid Scheduling Testing Environment

Abstract

With the phenomenal growth of the Internet and the advancement of computing hardware, grid architectures have been developed to exploit idle cycles in large networks of computational resources. One key aim of resource management (scheduling) schemes is to find mappings of incoming workload to machines within the grid to maximize the output. The first contribution of this thesis is the construction of a tool to aid researchers in testing and improving scheduling schemes, namely the McMaster Grid Scheduling Testing Environment (MGST). The Linear Programming Based Affinity Scheduling Scheme (LPAS_DG) was introduced by researchers at McMaster, and simulation results have been promising in suggesting that this scheduling scheme outperforms other schemes when there is high system heterogeneity and is competitive under lower levels of heterogeneity. The second contribution of this research is providing suggestions to improve this scheme, based on the results of experiments where the LPAS_DG scheme was actually deployed on the MGST test bed.