Inference on BitTorrent Mainline DHT from Network Evolution Time Series

Abstract. Network size is a fundamental statistic for a peer-to-peer system and is generally considered to contain valuable information for analyzing system dynamics. BitTorrent Mainline DHT, as the largest peer-to-peer system today, has been attracting more and more attention in the research community because it provides us an excellent opportu-nity to study the global user behavior in a content distribution system. However, most existing work only considers the metric by itself and does not explore what features could be extracted from this fundamental met-ric. What’s more, the similarity in the usage pattern of Mainline DHT has not been well studied in a quantitative way and how to cluster them remains as an open question. In this paper, we apply frequency-based feature extraction to such time series data. By using the proposed al-gorithm, our system successfully discovers and clusters the countries of similar user behavior and captures the anomalies like Sybil attacks and other real-world events with high accuracy. In addition, our result on Mainline DHT provides an interesting and alternative view the on global usage pattern in BitTorrent system