Enhanced P2P Content-Addressable Network And The Discussion Of Its Implementation Under The NGI Environment

The second generation of P2P based on Distributed Hash Table (DHT) is the newest progress and the hot research spot in this field. Content-Addressable Network (CAN) is a typical system among the second generation structured P2P systems. It organizes its data in virtual d-dimensions coordinate space by the use of DHT, thus it provides the ability of effectively locating resources according to the hash keys, the ability of fault-tolerant, and scalability. In addition, the inventor of CAN discussed the application layer multicast algorithm in CAN, and proved that this algorithm is especially effective in avoiding message reduplication. But the most DHTs, including CAN, still have two main problems: Firstly, they are lack of the ability to do effective content search. Key oriented DHTs search has already been aware of the exact search goal "specific key" before search, but it is impossible to content search to determine the exact search goal before the search process, because it needs to select the match results according to the similarity between the search goal and resources. Secondly, maintaining the network structure is a considerable cost. Although the DHTs need few network resources to query a resource, it needs many efforts to guarantee the accuracy of route table when nodes join in or depart from the network, especially in highly dynamic P2P network.From two angles: network behavior statistics and the algorithm optimization, the author of this thesis analyzes the two problems mentioned above and provides an enhanced CAN solution, which is called "Grid-CAN", based on content cluster network: Firstly, according to the network power law model as well as existing statistics of network resources distribution, the minority top content providers and responders, which are called "cluster peers", in the P2P network are organized by Grid-CAN into a virtual d-dimensions CAN structure. Thus the Grid-CAN can significantly reduce the structure maintance expense and still remain full abilty of content provide and response; Secondly, according to the small-world network model, Grid-CAN attaches small-world character into CAN, and reduces the average hops needed to flood a message among the whole network from o(N^1/2) to O(log²N). Forthmore, this Grid-CAN multicast algorithm is employed to do scable content search, which is absent in most of DHTs; Thirdly, according to the quantity statistics of different messages transferred in P2P network, Grid-CAN proposed a pipeline message transmission mechanism. It can not only eliminate the use of Ping-Pong messages (about 63% of total messages), but also increase the efficiency of network resource for message transmission.In order to evaluate the feasibility of Grid-CAN under the environment of next generation Internet (NGI), the author analysis the implementation of Grid-CAN as well as its support for two important next generation Internet technologies: Grid and IPv6. In order to evaluate the performance improvement of Grid-CAN, the three main algirthms of Grid-CAN: peer rank algorithm, content search algorithm, and pipleline message transmission algorithm are analyzed in detail and evalutated by simualation experiments.