Abstract
Peer-to-peer (P2P) networks attract attentions worldwide with its great success in file sharing networks (such as Napster, Gnutella, Freenet, BitTorren, Kazaa, and JXTA). As a major design pattern for future systems opposite to the traditional client-server paradigm, research on P2P networks is extremely important and could possibly radically alter the way of day-to-day use of computer systems. Efficient resource discovery remains a fundamental problem for large-scale P2P networks. In contrast to P2P networks, people in social networks can directly contact some acquaintances that potentially have knowledge about the resources they are looking for. Similarly to social networks where people are connected by their social relationships, two autonomous peer nodes can be connected in unstructured P2P networks if users in those nodes are interested in each other's data. The similarity between P2P networks and social networks, where peer nodes can be considered as people and connections can be considered as relationships, leads us to believe that theories of social networks are useful for improving the performance of resource discovery in P2P networks. In this thesis, we present Small World Architecture for peer-to-peer Networks (SWAN) and Efficient Social-Like Peer-to-peer (ESLP) model for resource discovery in P2P networks by organising peer nodes into a social peer-to-peer network either intentionally or spontaneously. SWAN is first presented in this thesis. In SWAN, each node keeps a list of neighbouring nodes in the same peer group, and peer groups are connected by a small number of inter-group links. Not every peer node needs to be connected to remote groups, but every peer node can easily find out which peer nodes have external connections to a specific peer group in SWAN. However, similarly to previous community-based P2P file-sharing systems, SWAN is based on the same concept of clustering peer nodes intentionally into peer groups, which needs extra communication overhead to maintain a hierarchical group structure in a highly dynamic and distributed environment. Addressing this issue, an Efficient Social-Like Peer-to-peer (ESLP) model is presented in this thesis to self-organise autonomous peer nodes on unstructured P2P networks by mimicking different human behaviours in social networks. In the ESLP model, queries are preferentially forwarded to the peer nodes that are more likely to have the requested resources or potentially have the knowledge on whom has the requested resources. Unlike community-based P2P models, we do not intend to create and maintain peer groups or communities consciously. In contrast, each peer node connects to other peer nodes with the same interests spontaneously by the results of daily searches. ESLP has been simulated in a dynamic environment with a growing number of peer nodes. From the experimental results, ESLP achieves better performance than existing methods, such as NeuroGrid.