Abstract
Peer-to-peer (P2P) networks have attracted significant attention due to their advantages on service availability, fault tolerance, self-organization, and provision of anonymity. P2P networks are built as overlays on top of physical IP network. However, peers randomly join and leave a P2P network resulting in topology mismatch between P2P overlay and physical IP network, which degrades the performance of P2P systems due to redundant and duplicate messages are transmitted over physical IP links. For matching the P2P overlay and the physical IP network, the location information of peers should be taken into account on selecting neighbors. This thesis proposed a two-tier locality-aware P2P architecture to alleviate the problem of topology mismatch in the P2P network by employing location information of peers to group nearby peers into clusters, which are led by superpeers. The upper tier is a Pastry [6] overlay constructed by superpeers, and the lower tier consists of clusters formed by normal peers that only connect to nearby superpeers. The simulation result shows that our proposed two-tier locality-aware P2P architecture could effectively improve the performance of P2P networks in term of average neighbor distance and search latency.