How low can you go: Balancing performance with anonymity in Tor

Tor is one of the most popular anonymity systems in use today, in part because of its design goal of providing high performance. This has motivated research into performance enhancing modifications to Tor's circuit scheduling, congestion control, and bandwidth allocation mechanisms. This paper investigates the effects of these proposed modifications on attacks that rely on network measurements as a side channel. We introduce a new class of induced throttling attacks in this space that exploit performance enhancing mechanisms to artificially throttle a circuit. We show that these attacks can drastically reduce the set of probable entry guards on a circuit, in many cases uniquely identifying the entry guard. Comparing to existing attacks, we find that although most of the performance enhancing modifications improve the accuracy of network measurements, the effectiveness of the attacks is reduced in some cases by making the Tor network more homogeneous. We conclude with an analysis of the total reduction in anonymity that clients face due to each proposed mechanism.