Efficient Content Locating in Dynamic Peer-to-Peer Networks

Abstract

The Peer-to-Peer (P2P) computing model has recently been recognized as a more natural and flexible approach to sharing resources. However, a fundamental issue, content locating (or content search) in P2P-based applications has not yet been successfully resolved. This thesis documents the design and implementation of a novel architecture, which combines the advantages of cluster infrastructure and Super-Peer overlay to address the scalability, robustness and efficiency issues in existing unstructured P2P systems. The first component of our architecture is called the TBCP model, which constructs a set of interconnected clusters, where each cluster forms a bounded-depth tree, and each peer node acts as a tree leaf. The duties of maintaining cluster topologies and providing search services are separated and re-balanced to address heterogeneity. The COOL model, the second component of our architecture, then constructs light-weight interconnected overlay networks, by following simple classification and mapping rules. Peers are connected in both vertical direction (within the cluster domain) and horizontal direction (within the Super-Peer overlay domain). We also propose two search algorithms, known as cluster search and overlay search, which works seamlessly with our new models to provide efficient and low-cost content locating services. Our experiments and analysis prove that our new architecture achieves significantly better scalability and efficiency than a basic unstructured model.