BitTorrent traffic obfuscation: A chase towards semantic traffic identification

With the beginning of the 21st century emerging peer-to-peer networks ushered in a new era of large scale media exchange. Faced with ever increasing volumes of traffic, legal threats by copyright holders, and QoS demands of customers, network service providers are urged to apply traffic classification and shaping techniques. These systems usually are highly integrated to satisfy the harsh restrictions present in network infrastructure. They require constant maintenance and updates. Additionally, they have legal issues and violate both the net neutrality and end-to-end principles. On the other hand, clients see their freedom and privacy attacked. As a result, users, application programmers, and even commercial service providers laboriously strive to hide their interests and circumvent classification techniques. In this user vs. ISP war, the user side has a clear edge. While changing the network infrastructure is by nature very complex, and only slowly reacts to new conditions, updating and distributing software between users is easy and practically instantaneous. In this paper we discuss how state-of-the-art traffic classification systems can be circumvented with little effort. We present a new obfuscation extension to the BitTorrent protocol that allows signature free handshaking. The extension requires no changes to the infrastructure and is fully backwards compatible. With only little change to client software, contemporary classification techniques are rendered ineffective. We argue, that future traffic classification must not rely on restricted local syntax information but instead must exploit global communication patterns and protocol semantics in order to be able to keep pace with rapid application and protocol changes.