Degree Name

Masters by Research


School of Information Technology and Computer Science - Faculty of Informatics


Grid computing as an emerging technology has made great achievements in scientific computation. Leveraged by other technologies, such as cluster computing and web services, Grid computing for the first time seamlessly integrated large numbers of interconnected computers and provided Internet-scale computing resource sharing, selection, and aggregation. The sheer numbers of desktop systems today make the potential advantages of interoperability between desktops and servers into a single Grid system quite compelling. However, these commodity systems exhibit significantly di_erent properties than conventional server-based Grid systems. They are usually highly autonomous and heterogeneous systems, and their availability varies from time to time. We call such an environment an open environment. This thesis aims at bridging the gap between conventional Grid computing and its potential application in open environments by proposing an agent-based peer-to-peer Grid computing architecture, whilst also providing reasonable compatibility and interoperability with conventional Grid systems and clients. We introduce developments in Grid computing and highlight the targeted research questions concerning Grid computing in open environments. Using these questions as a basis, we review the architecture of the conventional computing Grid and related standards. We indicate that the conventional Grid has five problems, which are barriers to the deployment and application of the Grid in an open environment. Aiming at solving these problems, we propose a hybrid solution, which is a combined solution that employs both client/server computing architecture and peer-to-peer computing architecture. This solution abandons conventional super-local Grid architecture, and is more e_cient, flexible and robust in open environments. We also introduce a multi-purpose task model to handle state persistence and provide help v for task decomposition. Furthermore, we employ multi-agent technology to construct the underlying components of our Grid architecture, which brings flexibility and robustness. Based on the hybrid solution, we develop the architecture to a pure peer-to-peer architecture. In the new solution, we make improvements to the task model so that it provides additional support to task decomposition and inter-task communication in a transparent manner. We develop two frameworks for message passing and routing, and for resource management respectively. These frameworks, together with various intelligent and evolving mechanisms, promote the new adaptability and performance to a higher level. We widely adopt Web Services and other Grid standards in both solutions to maintain compatibility and interoperability with existing Grid systems and clients. Finally, we discuss the remaining problems with Grid computing in open environments, and outline potential research directions. In summary, we show that Grid computing architecture, integrating peer-to-peer computing and multi-agent technologies, presents good scalability, e_ciency, flexibility, and robustness for Grid computing in open environments in comparison with conventional Grid computing architecture.



Unless otherwise indicated, the views expressed in this thesis are those of the author and do not necessarily represent the views of the University of Wollongong.