Open Collections

Abstract

Peer-to-peer (P2P) computing has become increasingly popular in recent years. It offers many attractive features, such as self-organization, load-balancing, availability, fault tolerance, and anonymity. However, it also faces some serious challenges. In this thesis, we propose an Efficient Clustered Super-Peer P2P architecture (ECSP) to overcome the scalability and efficiency problems of existing unstructured P2P systems, using a semi-centralized hierarchical structure: With ECSP, peers are grouped into clusters according to their topological proximity, and super-peers are selected from regular peers to act as cluster leaders and service providers. These super-peers are also connected to each other, forming a backbone overlay network operating as a distinct, yet integrated, application. To maintain the dynamically adaptive overlay network and to manage the routing on it, we propose an application level broadcasting protocol: Efa. Applying only a small amount of information about the topology of a network, Efa is as simple as flooding, a conventional method used in unstructured P2P systems. By eliminating many duplicated messages, Efa is much more efficient and scalable than flooding, and furthermore, it is completely decentralized and self-organized. Our experimental results prove that ESCP architecture, combined with the super-peer backbone protocol, can generate impressive levels of performance and scalability.