Selective Grouping Algorithm For Low Latency Anonymous Systems

Abstract

Low latency anonymous communications are prone to timing analysis attacks. It is a technique by which the adversary can de-anonymize the user by correlating packet timing patterns. A recent proposal to stop these attacks is called Dependent Link padding. However, it creates high dummy packets overhead in the network. In this work we propose selective grouping, a padding scheme that protects users in an anonymity system from those attacks with minimal overhead. The aim is to decrease overhead by dividing users in different groups while maintaining good anonymity. The key idea of our approach is to group clients with similar timing patterns together by providing a strict delay bound. We ran simulation experiments to test the effectiveness of these techniques and to measure the amount of extra network congestion. We have also statistically analyzed bursty traffic in the network by using the mean and standard deviation of inter packet delays over a fixed duration. The result of bursty traffic analysis added one more dimension to the count of packets for grouping clients efficiently. To analyze anonymity, we ran a statistical disclosure attack against our selective grouping defense. We performed extensive sets of experiments to find a threshold value at which selective grouping achieves good profiling without adding excess dummy packets. We show that selective grouping is very effective at resisting timing analysis attacks and are still able to provide good anonymity with minimal overhead added to the network.