Measuring web feature impacts in Peer-to-Peer file sharing systems

In Peer-to-Peer (P2P) file sharing systems, the presentation of resource attributes organized in different web features can influence users’ selection. However, this has been only qualitatively speculated without concrete analysis. In this paper, we conduct extensive quantitative measurements on the impacts of these attributes by crawling web pages of a BitTorrent site, “5QZone”. The measurement lasts for 31 days, and 168,610 records containing 11,228 distinct resources have been collected. We further compare it with one-day commercial measurement data from a hybrid file sharing site, “Xunlei”, which contains 5,473,283 resources for 5,156,696 participating users. The finding is twofold. On one hand, it confirms the above qualitative speculation; on the other hand, it shows more significant results: (1) with the highlight feature on popular items, the downloads of each resource yield to a long-tail distribution, deviated from Zipf Law; (2) publications with attracting titles disseminate substantially faster than others; (3) publisher authority feature has limited influence on depressing low quality resources; (4) other features such as presenting resource categories and sizes also affect user behavior. We further demonstrate the implications of the web feature impact for system design and potential attackers.