| Current low-latency anonymous communication networks suffer from both relatively poor performance and long-standing security vulnerabilities. In order to achieve widespread adoption, both of these problems must be addressed and usability must be improved. Unfortunately, there is a natural tradeoff between performance and anonymity such that improvements to one tend to harm the other. In order to address these issues, this thesis re-examines the process of path selection in low-latency communication networks from the ground up.;A novel peer-to-peer bandwidth evaluation protocol, called EigenSpeed, is presented and shown to be accurate and secure even in the face of a large group of colluding attackers. Systems using this protocol in place of self-reported bandwidth information are no longer vulnerable to several classes of well-known attacks. The algorithm is further shown to produce better predictions of achieved performance than do current self-reporting techniques.;A new router-selection algorithm for low-latency anonymous communication networks, called Tunable Tor, is introduced, which allows users to customize the inherent performance-anonymity tradeoff to suit their preferences. Additional changes remove the need for artificially imposed bandwidth caps. This algorithm enables significantly higher throughput, higher anonymity, or smaller improvements in both. The algorithm is also analyzed and shown to be resistant to leaking information about the expressed anonymity preferences.;Finally, a method of further improving throughput in low-latency anonymous communication networks through the use of multipath routing is introduced. Practical techniques are presented allowing implementation in current networks without fundamental architectural changes. Analysis of the performance of multipath routing in combination with the previously presented algorithms shows a twenty-fold improvement in some performance metrics while maintaining anonymity at or above previous levels. |
