Research And Implementation On Content Distribution Structure Based On Kademlia

Nowadays, various information bombard people's daily life. And everyone is expecting higher quality of internet service and more fast information fetch. Technically speaking, the two current prevalent technologies are Content Delivery Network (CDN) and P2P.The core of CDN is to layout servers on the network edge, which can release content to the "edge" which is nearest peer to the user. The user can get needed information from the nearest nodes. This technology brings fast response, decrease of server load and increase of network devices utilization. To the nature the core is still based on C/S structure, configure the high cost servers in order to the effect of one step away and associated with region of the control. The cost for maintenance and expansion is very high. It can not take full advantage of a large number network edge resource. P2P technology breaks the traditional C/S model. Every peer in it is both customer and server. Peers get resources from other peers and provide resources for others at the same time. P2P proposed totally different ways of delivering contents. The core of P2P is to fully utilize user resource for peer-to-peer transmission. Fast file exchange without rely on server was implemented by the multi-point transmission mechanism and design of decentralized. P2P technology is applied in network applications widely, especially Kademlia protocol is more widely because of its efficiency and stability.We think of the "one step" method in the CDN, study the locality information of P2P node, proposed a cross layer locality-awareness and interest-focusing Kademlia (LIKad) structure. The structure contains source server layer, index_peer layer and autonomous peers layer. The LIKad structure consists of different index_peers in different zones. We improve the search algorism which can match nodes' logical topology and physical topology and increase the efficiency of searching and sharing resource. Autonomous peers consist of common customers. Nodes in the same zone and having same interest share resource first. Then they vote for powerful node to be their index, which is in charge of managing and organizing nodes in its zone. Experiments show this structure has low latency,large scalability, high performance in query, load balance. It's very suitable for dynamic network when the scale is very large.