An Adaptive System Research And Implementation Based On Kademlia Protocol

P2P technologies are characterized by non-centralization, scalability, high-performance, robustness et al. It has been widely used in practice. According to whether overlay network maintains a specific topological structure, P2P technologies are divided into two classes:structural and non-structural. Flooding searching was used in non-structural networks, which cannot guaranty the robustness and scalability of the systems. Structural networks use Distributed Hash Table (DHT) for data storage, indexing and maintenance and have a good performance for searching. The Kademlia(KAD) protocol is an implementation of DHT, however, it uses a special XOR algorithm in distance measure and establishes a new topological structure. Compared to other algorithms, this XOR algorithm efficiently enhances the route tracing speed. However, topological structure of the physical networks below the overlay is not considered when building the overlays in these implementations. The mismatching of the overlay network and physical network leads to excessive network latency. This thesis makes analysis and research on matching the topology of physical network and that of overlay network. The main contents are as following:1. I make a classification of current P2P networks. I investigate some representative implementations of decentralized structured system and make a comparison of their performance.2. I make a deep research on the widely-used Kademlia protocol and present its merits and limits.3. I make a research on the mismatch problem of physical network and overlay network and propose a method based on neighbor information exchanging to solve this problem.4. Based on the above, an aKad (adapted Kademlia) protocol is designed to make the matching of the two networks at different layers adaptively.5. Simulations of the improved protocol are made on the platform of OMNeT+++ INET+OVERSIM. The results show that, compared to the original Kad protocol, the improved method in this dissertation is actually improving the matching rate of physical and overlay networks and reduce the overall delay of network.