An Empirical Evaluation of Relay Selection in Tor

While Tor is the most popular low-latency anonymity network in use today, Tor suffers from a variety of performance problems that continue to inhibit its wide scale adoption. One reason why Tor is slow is due to the manner in which clients select Tor relays. There have been a number of recent proposals for modifying Tor’s relay selection algorithm, often to achieve improved bandwidth, latency, and/or anonymity. This paper explores the anonymity and performance trade-offs of the proposed relay selection techniques using highly accurate topological models that capture the actual Tor network’s autonomous system (AS) boundaries, points-of-presence, inter-relay latencies, and relay performance characteristics. Using realistic network models, we conduct a wholenetwork evaluation with varying traffic workloads to understand the potential performance benefits of a comprehensive set of relay selection proposals from the Tor literature. We also quantify the anonymity properties of each approach using our network model in combination with simulations fueled by data from the live Tor network.