Finding Peer-To-Peer File-Sharing Using Coarse Network Behaviors?

This paper was presented at ESORICS 2006, which took place in Hamburg, Germany, in September 2006

A user who wants to use a service forbidden by their site's usage policy can masquerade their packets in order to evade detection. One masquerade technique sends prohibited traffic on TCP ports commonly used by permitted services, such as port 80. Users who hide their traffic in this way pose a special challenge, since filtering by port number risks interfering with legitimate services using the same port. We propose a set of tests for identifying masqueraded peer-to-peer file-sharing based on traffic summaries (flows). Our approach is based on the hypothesis that these applications have observable behavior that can be differentiated without relying on deep packet examination. We develop tests for these behaviors that, when combined, provide an accurate method for identifying these masqueraded services without relying on payload or port number. We test this approach by demonstrating that our integrated detection mechanism can identify BitTorrent with a 72% true positive rate and virtually no observed false positives in control services (FTP-Data, HTTP, SMTP).