Study On P2P File Distribution With Network Coding

File distribution is to distribute source files to a lot of end users. It is involved in a type of applications, such as file sharing, living streaming, and etc. As the development of Internet and the growth of the amount of network users, the quality requirement of file distribution from users also increases. P2P makes use of computing and uploading resources of end users, and improves the performance of file distribution. Network coding is a new research area in information theory. With network coding, intermediate nodes between source and destination(s) encode the incoming packets into some new ones to the outgoing links while transmitting, and the original content is decoded at the destination(s). With P2P and network coding, the performance of file distribution systems can be improved.The main contributions of this thesis are:1) We propose a sparse linear coding algorithm, and analyze the rate of its successful decoding based on Cooper's results on sparse matrixes. With sparse coding,the complexity of network coding can be greatly reduced from O(n) to O(log n),where n is the block number of the file to be encoded. In experiments, the encoding speed of sparse coding reaches 7MB/s, which is high enough to catch uploading speed, and keeps a high rate of successful decoding.2) We implement a P2P file distribution system with sparse linear coding (SNC), and with the same architecture, we also implement a system without network coding (LRF), which uses local-rarest-first scheme to schedule blocks. Experimental results on a LAN testbed show that SNC is better than LRF in terms of average downloading time, total distribution time and the whole throughput. Averagely, sparse coding reduces about 10-15% downloading time.3) We introduce a pre-checking mechanism in order to avoid linearly dependent blocks being transmitted. With the same architecture, we implement a P2P file distribution system with sparse coding and pre-checking (SmartCode), a system with sparse coding but without pre-checking (Ecode) and a system without network coding (LRF). We carry out wide-area network experiments on PlanetLab. The experimental results show that SmartCode performs better than Ecode and LRF both in the average downloading time and robustness. Averagely, SmartCode reduces 7% downloading time compared with Ecode and 11% downloading time compared with LRF. In some experiments about robustness under the assumption that the peers can leave the system, no peer completes the downloading with LRF, and a few peers complete the downloading with Ecode, but over 75% peers complete the downloading with SmartCode.