Abstract

Grid computing is an established concept aimed at providing high performance computing resources to end users around the world. The resources are generally in the form of powerful super computers or workstation clusters. The major challenge comes in the form of a middle-tier implementation that will help successfully route tasks from their originating users to a suitable resource provider.

The aim of this study is therefore to assess the flow of tasks through a standard three tier grid computing architecture. The study will account for agent sub-systems in the form of users, brokers and resource providers.

The study will provide new or novice users in the field of high performance computing with an educational tool to demonstrate the practical workings of a simple system. The educational tool will provide a clear user interface to allow users to follow a task through all the stages from creation to completion.

The architecture will model and demonstrate the dependencies between layers through the use of manual investigation and simplistic statistics gathering. An investigation will determine how changes in the task requirements and resource parameters (such as load, hardware configuration, etc) effect the overall task distribution within the macro environment of the model.

Due to the complex nature of the research field it is intended that the study should only focus on a sub-set of the full features found in a grid computing architecture. The results obtained from the implemented feature set will then allow extrapolation of trends and behaviours affecting some of the more advanced features, such as Quality of Service agreements, network issues, security and inaccurate data.

LSD modelling will be utilised to further demonstrate the individual agent perspectives and aid in the creation of a realistic dependency model for each agent sub-system.

It is anticipated the study will highlight some of the issues involved in the passing and assignment of pre-defined, parameterised tasks. As well as this issues relating to interaction between the various agents will also become apparent.