XunleiProbe: A Sensitive and Accurate Probing on a Large-Scale P2SP System

Xunlei [1] is a new P2P content distribution system which is popular in China. It composes traditional HTTP/FTP downloading and P2P content distribution features which attract many people including researchers. Xunlei´s network is Bit Torrent-like and the measurement is more difficult than other P2P networks [2]. There are many constrains on Xunlei tracker, so we can not obtain information from tracker easily. Besides this, the measurement will encounter some challenges that skew the results. Most of previous works on Bit Torrent system are based on tracker logs. However, we can not obtain tracker logs of Xunlei. As far as we know, there is no proposal about precise and detailed measurement method that probes the network directly in this area. Face to the challenges, we analyze the constrains that appear in most Bit Torrent-like systems and propose a stratified random selection model to describe the behavior of tracker. Based on the model, we design a measurement tool called XunleiProbe. With the help of our solutions that increase the accuracy of our results, we measure a popular swarm for about 22 hours. The results show that the average peer coverage of our tool can reach about 93%.