Correlation attacks on the Tor network

Internet privacy has become one of the most important concepts of our time. People are worried about the surveillance of their digital communications and are in need of reliable tools to preserve their anonymity as well as the privacy and the integrity of their communications. Researches have been made to design and set up such reliable systems and The Onion Router (Tor) is one of them. Like many other systems, Tor is far from being perfect and may be subject to traffic correlation attacks. Such attacks can be carried out using a variety of methods and approaches. In this context, we develop the threat-model of an adversary supervising Tor relays and carrying correlation attacks using Tor log files. Three different correlation methods are implemented and their performances are evaluated on reduced-size Tor networks simulated with the Shadow simulation tool. The performances are evaluated considering different client models and reveal what method is more suitable for the adversary. Furthermore, the document develops the procedure an adversary could follow in order to carry out the attacks in a realistic environment. The results we obtained through simulations suggest that an adversary could conveniently use the Tor log files generated by Tor controller to perform successful end-to-end correlation attacks against the different client models. Furthermore, the results also suggest that among the three correlation methods that were implemented, one was surely more reliable than the other. The results eventually illustrate that the most reliable correlation method performs well against every client models that were simulated. This suggests that Tor users can either do some instant-messaging, web-surfing or even download some large files, they are almost under the same level of threat.