Research On Resource Location Technologies In Unstructured Peer-to-peer Network

The peer-to-peer (P2P) network as a new computing model has been used by more and more individuals, companies, governments and organizations. P2P networks allow the computing participants to share their resources through the Internet. Because of the inherent characteristics, such as uncertainty, distribution and openness, P2P networks are facing serious challenges on the increasing network scale and application popularization. It has become increasingly difficult to efficiently manage the huge resources that are stored in jointed peer node respectively.Resource location is an important issue of P2P network and has been widely concerned by researchers. It is intended to provide service consumer with the services of quick access to target resources for the users through building resource location indices. In this thesis, we survey the state of the art in the P2P networks resource location studies, review the past progresses and important achievements in this area, and summarize the key technologies and difficulties of resource location in unstructured P2P networks. In the background of the Internet environment, this thesis focuses on some key issues of the resource location in large scale unstructured P2P networks from optimizing overlay network topology, enhancing the fault tolerance of models and improving the efficiency of resource location algorithms. The contributions are as follows,(1) A Locality Aware distributed Spanning Tree (LAST) model was presented to resolve the topology mismatch problem in unstructured P2P networks. The latency of underlying network was defined as the distance between nodes in the LAST model, and where the metric space and distance were defined and the rules of determining adjacent nodes and selecting nearby groups were also proposed. Furthermore, the logic definition of the LAST model was described through the (a, b) coding tree. In the LAST overlay network, nodes join and leave the model according to the node management algorithms. The mathematic analysis and simulation results show that the LAST model has the Small World characteristic; Comparing to the distributed spanning tree (DST) model, the LAST model has better self-adaptive and load balance by means of the node management algorithms with logarithmic time complexity, and the locality-aware capability reduces almost 60 percent of the average distance and 40 percent of the average latency.(2) A Fault Tolerance enhanced LAST (FT-LAST) model was proposed to improve the robustness of unstructured P2P networks. Based on the research on the LAST overlay network, the Relationship Vector Similarity (RVS) of nodes was defined, after that the rule of Rvs-Based Representative Selection (RBRS) was also described. Without increasing the number of redundant connections and resource replicas, the fault tolerance of that model was proactively enhanced with smaller overhead. The simulation results show that the probability of critical nodes appearing in the FT-LAST model is reduced and the model remains connected while the random failure probability less than 65%. The mathematical analysis further describes the fault tolerance of the FT-LAST model with the adversarial failure, implying that only O(f / (log N ? log f )) nodes can be separated from N nodes by f failures and therefore its performance is better than other similar models.(3) A Search-Radius Limited (SRL) resource location algorithm was presented to resolve the high search-costs and low efficiency problems of flooding searches in unstructured P2P networks. In the FT-LAST model, the definition of search-radius was described, and the network overhead of flooding searches was reduced by limiting the search-radius of the SRL algorithm. By providing four search strategies of the SRL algorithm, the system flexibility was promoted in the face of different urgency tasks and user levels. A method to determine the value of search-radius on the premise of ensuring the satisfaction at search results was proposed. The efficiency of the SRL algorithm was improved by restricting the spread of search messages. The mathematical analysis proves that the time complexity of the SRL algorithm is constant order. By comparing the simulation results of the four strategies with the DST′s, the performance of the SRL algorithm is better than others.(4) In this thesis, the author designed and implemented a P2P application service platform FlasWire based on the FT-LAST model and LimeWire, which is the most popular open source client supporting the Gnutella network. By expending the Gnutella protocol modules of LimeWire, FlasWire nodes can be organized into the FT-LAST overlay network which provides the locality-aware capability and strong fault tolerance. Through adding the SRL-based resource location module, FlasWire improves the resource location efficiency and reduces the network overhead of flooding searches. As all the designs and implementations are based on open architectures and protocols, FlasWire facilitate the third-party platform expending and rapid secondary developments.