Archived: 11 February 2003
University of Adelaide Honours thesis, October 2002.
Supervisors: Andrew Wendelborn, Paul Coddington and Kevin Maciunas
Grid computing is a paradigm of distributed computing in which heterogeneous nodes are spread over wide areas and are connected by different bandwidth connections. The intention of grid computing is to make available under-utilised or specialised compute resources to users who wish to utilise them. The Globus Toolkit has been developed as part of a research project into grid computing. The toolkit provides components known as grid middleware. Middleware is software that runs on the nodes of a grid that allows users to gain access to other resources on the grid, it does this by providing interfaces and protocols that smooth over problems associated with executing processes in a distributed environment.
An important aspect of grid middleware is the provision of a resource management architecture. The current implementation of Globus contains a hierarchical directory that users can query to determine what resources are available and what their current levels of usage are. The toolkit also provides servers that run on nodes to which users can submit requests to run jobs. There is no provision in the toolkit of a resource broker that for managing the process of determining available resources, and deciding those which would be appropriate for an application.
Presented here is the outline of the development of such a component that provides resources selection services to developers of grid applications. The GERBIL resource broker takes a grid applications resource requirements, as expressed through RSL, and matches those requirements against available grid resources. RSL provides a powerful and expressive tool for specifying resource requirement s associated with a grid application. Results are presented that demonstrate how GERBIL interprets these specificatio ns to provide an effective tool in the development of grid applications.
Postscript version (gzip compressed)