John Heidemann

Inherent Behaviors for On-line Detection of Peer-to-Peer File Sharing

TitleInherent Behaviors for On-line Detection of Peer-to-Peer File Sharing
Publication TypeConference Paper
Year of Publication2007
AuthorsG. Bartlett, J. Heidemann, and C. Papadopoulos
Date Publishedmay
Conference LocationAnchorage, Alaska, USA

Blind techniques to detect network applications–-approaches that do not consider packet contents–-are increasingly desirable because they have fewer legal and privacy concerns, and they can be robust to application changes and intentional cloaking. In this paper we identify several behaviors that are \emphinherent to peer-to-peer (P2P) traffic and demonstrate that they can detect both BitTorrent and Gnutella hosts using only packet header and timing information. We identify three basic behaviors: failed connections, the ratio of incoming and outgoing connections, and the use of unprivileged ports. We show that while individual behaviors are sometimes effective, they work best when used together. We quantify the effectiveness of our approach using two day-long traces, from 2005 and 2006, showing that they are quite accurate: BitTorrent hosts are detected with an 83% true positive rate and only a 2% false positive rate, and Gnutella hosts with a 75% true positive rate and a 4% false postivie rate. Our system is suitable for on-line use, with 75% of BitTorrent hosts detected in less than 10 minutes of trace data.