Research On Optimizing Technology Of Peer-to-Peer Overlay Network Topology

Peer-to-peer (P2P), an emerging model that relies primarily on computing power and bandwidth of participating peers instead of centralized infrastructures, is scalable, robust, and can be easily deployed. P2P computing is a promising architecture for distributed services over the Internet, and become the hottest topic of network computing in recent years.The topology of a P2P overlay network is the most important character which influences the performance of a P2P system. First, as a P2P system is highly dynamic, the resilience of overlay network is of great importance, so as to make sure that the system will not crash due to the peers' leaving, and is not vulnerable to any attacks. Second, search is the most important operation in P2P environments. The efficiency of searching would determine the performance and scalability of a P2P system. The topology optimization is one of the major approaches for improving search efficiency. In order to improve the system's reliability and search efficiency, we study the topology optimization in depth. The major contributions of this research are as follows.We propose a distributed cut vertex discovery and neutralization algorithm in P2P environment, which depends on local knowledge to enable every single node be able to determine whether it is a cut vertex or not, and adapt effective methods to clear up its badness to the network. We prove the correctness of this algorithm and evaluate the performance of our design through trace driven simulations. Our design is self adaptive, high veracity and with minimal overhead, and it can discover all cut vertices with very small proportion of the peers wrongly treated as cut vertex in static network. It is also highly accurate in active networks.We propose a distributed mechanism which efficiently detects the vertex cutset with 1 or 2 vertices and neutralize them into normal nodes, by smartly counting the number of disjoint paths between the neighbors of the selected peers. We evaluate the performance of our design through trace driven simulations. The results show that our algorithm greatly improves the reliability of an overlay network upon the failure of vertex cutset.We also propose a topology optimization algorithm ILS by link selection in Gnutella-like systems, to address the search efficiency problem by exploiting the principle of interest-based locality. We prove a model to evaluate the similarity of peers and value of the connections. Peers build new connections with others which have similar interests and disconnect those with minimal values. Our simulation study shows that ILS can significantly reduce the reply path lengths, achieving high search success rate, as well as reducing the total communication cost in unstructured P2P systems.We propose a novel model for Data Grid Information Service using a deterministic P2P based hypercube topology, which realizes the keyword-based searching in the cost of O(logN). Furthermore, we propose a transposition algorithm to optimize the overlay network's topology according to the access statistics between peers, making the peers neighbors when they often visit each other. The simulation and analysis shows that our approach could sufficiently utilize the special character of Date Grid Information Service, and guarantee the QoS and scalability of the system effectively. Also, the transposition algorithm significantly improves search efficiency.