Archived: 11 February 2003
University of Adelaide Masters by coursework thesis, June 2002.
Supervisors: Paul Coddington, Andrew Wendelborn and Kevin Maciunas
A computational grid is a hardware and software infrastructure for distributed computing. Recently, it has caught considerable attention, even from big companies like IBM, Sun, Compaq. Drives come from the increased network bandwidth, the more powerful computers, and the growing workload demands. However, many challenges are in the way, when computing grids try extending beyond their home organizations, such as the concern of security, the discovery and utilization of resource, the diverse policies, the performance uncertainties, and so on.
This thesis focuses on the construction and resource management in the context of a campus grid, a wide-area computing grid across a university campus consisting of multiple administrative domains. The software for building a campus grid includes Cluster Management Software (CMS), which has been applied widely to local grids, and the Globus Toolkit, which is addressing the issues of wide-area networks and is undergoing rapid development.
The paper will first give a brief study on a number of CMSs and Globus Toolkit. Then based on the differences and similarities among CMSs and Globus, I will give a blueprint for the construction and the resource management of a campus grid at University of Adelaide. Finally, for improvement of performance and the quality of service, I will give some suggestions about mechanisms by which Globus can be better integrated with CMSs.