同儕網路 (Peer-to-Peer；P2P) 近幾年來被大量的應用在網路服務上，而一個同儕系統如果具有位置知覺能力的話將會大幅增加系統效能。所以本篇論文提出利用層級架構 (hierarchical structure)來達成位置知覺並且將使用者依興趣分群的混合式P2P系統。在本系統中，使用者會依照位置區域和興趣分群並且利用超立方體結構 (hypercube structure) 來替使用者定址。整個系統從位置區域的觀點來看，系統將依實體位置相近的使用者分配在同一個區域之內，從興趣的觀點來看，系統將把使用者依興趣性質邏輯上的分成若干個興趣叢集 (Interest Cluster)，然後經由選舉出的超級節點 (super node) 來替其底下普通節點 (peer node)來進行跨區域與跨興趣的詢問，也就是說在本系統之下大部分的詢問訊息 (Query Message) 將只在鄰近的區域與同樣的興趣叢集之下發生，進而提升同儕網路的詢問與搜尋效能。我們還為系統運作設計了三種通訊協定，分別是節點加入、離開與搜尋通訊協定。最後我們利用 NS2 來模擬系統，並分別以點對點延遲與詢問所經過的 hop 數模擬距離以評估本系統的效能，模擬的結果顯示出在依時間順序加入系統的節點，假設其位置是均勻分散且網路節點相當多的情況之下本篇論文所提出的方法會大大的提升系統效能。 Peer-to-peer system (P2P system) is becoming a key technology and has been widely a platform for users to search and share file over the Internet. However P2P system tends to waste the network bandwidth because of its decentralized resource management. Consequently, efficient querying or searching becomes one of the important design issues in peer-to-peer system (P2P system). In this paper, we proposed a locality-awared hybrid P2P system with interest grouping. In this system, users are physically divided into several locality regions and logically grouped into several interest clusters. And we employ a hypercube structure to address users. Moreover, super nodes are elected to route the inter-region and inter-cluster query messages. That is the bulk of query messages will be routed in the same locality region and interest cluster. We also devised corresponding node Join, Leave and Search protocol. Finally, we evaluate the system performance of our approach by NS2 simulator. The experimental results show that if the number of participant nodes in the network are enough and even the nodes which the location are uniform distributed will join the system in time sequence, our approach will be outstanding. The results show that the query delay will be reduced to 1/6.63 with comparing to PeerCluster approach.
