Abstract

Grid computing is an emerging technology which has the potential to deliver high performance computing by facilitating access to geographically distributed computing resources and presenting them to the user as a single unified resource. It is a form of distributed or parallel computing which utilises agent technologies to automate activities and therefore hide complexities from users. An agent is viewed as a software entity that encapsulates a well defined functionality such as providing a particular service. Simulation environments which are able to accurately model Grid systems are crucial as they enable the operation of computational Grids to be examined in a repeatable manner. This would otherwise not be possible as due to the nature of Grid systems resources are often located across multiple administrative domain boundaries. Consequently it is impossible to gain access to these resources in order to recreate the exact system conditions which occurred in previous experiments. Simulators also play an important role in designing Grid systems and optimising existing systems. They allow the operation of a proposed design to be tested and improved before the actual system is built and enable the performance and configuration of existing systems to be examined and improved. Several metaphors have been proposed as a basis for Grid systems. This work is based on the computational economy paradigm and views Grids as economies of resources, one of the aims of this project is to examine how employing different economic models affects the performance of the overall Grid system. These economies consist of two basic types of participant, resource providers and resource consumers. Resource providers allow their resources to be accessed and are motivated by financial incentives, they require the highest possible return on their investment and therefore require their resources to be utilised as much as possible. Resource consumers require their computational jobs to be completed to satisfy their particular quality of service requirements, for example to satisfy budget or completion deadline constraints. A simulation environment would therefore enable the configurations which produce the highest overall throughput and achieve the best load balancing to be determined. It would also be possible to examine precisely where particular jobs are executed, which machines are over or under utilised and the possible effects if parts of the network were taken offline or the size of the Grid was increased. This Empirical Modelling (EM) project is therefore related closely to the field of Concurrent systems modelling. It is hoped that employing an EM approach will enable dependencies present within Grid systems to be more accurately modelled and so produce more realistic representations than has previously been possible using more traditional simulation techniques.