Kademlia 是一種分散式雜湊表 ( DHT : Distributed Hash Table ) 技術,作為 IP 之上的覆蓋網路,每個節點以 Kademlia 路由表,稱之為 K-bucket ,維護部份拓樸資訊,提供 P2P 網路資料存取能力。然而 IP 網路以位址為識別的點對點通訊機制,卻限制了 Kademlia 架構的擴展。 本研究以內容中心式命名資料網路 ( NDN : Named Data Network ) 取代底層 IP 路由,透過 NDN 以「名稱前綴」為識別的封包路由模式,以及「興趣」與「資料」兩種不同類型之封包,設計相應名稱前綴使 Kademlia 查找資料時能動態匹配節點,從路由層面有效解決 Kademlia 路由表因其節點紀錄規則,所形成群集現象導致的查找錯誤。進一步提出 Kademlia 雙向路由表、 逐跳式 Look up Method 等架構,降低延遲與路由成本。 此外,本研究提出一個基於邊緣運算的智慧商圈情境,並以商品推薦服務為主。在情境中,NDN 提供了網路內服務發現機制,契合 P2P 網路特性,Kademlia 則作為資料管理架構,有效解決 NDN 前綴氾濫問題,通過融合 NDN 與 Kademlia,建立具有高查找效率的分散式資料儲存網路,運行智慧商圈中的多樣化服務。
Kademlia uses a Distributed Hash Table (DHT) technique to form a P2P overlay network above IP. Each node maintains some topology information with Kademlia routing table, called K-bucket, and provides P2P network data access capabilities. However, the IP network's peer-to-peer communication mechanism limits the extension of the Kademlia architecture. In this study, a content-centric Named Data Network (NDN) is used as the underlying routing instead of IP. Routing in NDN is identified by the “name prefix”, and two types of packets (“interest”& “data”) are using for searching and reply. We modify Kademlia K-bucket to make good use of NDN in order to dynamically match node for searching data effectively. We also mitigate the search errors caused by the K-bucket clustering phenomenon. In addition, the Kademlia bidirectional routing table and hop-by-hop Look up Method are proposed to further reduce latency and routing costs. This study also proposes a smart business district scenario based on edge computing, with product recommendation as the main service, using the integration of NDN and Kademlia technologies. In the scenario, NDN provides a service discovery mechanism in the network,and Kademlia is used as a data management architecture to effectively solve the problem of NDN prefix flooding.