Improving stream correlation attacks on anonymous networks

The level of anonymity offered by low latency, interactive, anonymous networks is unknown. This paper implements correlation attacks on the deployed Tor network and a simulated Tor network under defined network conditions. The accuracy of the attacks act as a metric for the networks anonymity in the face of a passive adversary. From observation of the deployed Tor network, several techniques were developed to compensate for some of the modifications the Tor protocol induces in traffic. These techniques increase correlation accuracy by 10% to 40% for differing correlation functions. Almost 50% of traffic streams on the simulated network are identified immediately with 10% of experimental traffic on the real Tor network identified.