Shaping Network Topology For Privacy And Performance
Abstract
Anonymity systems provide privacy for Internet communication and are becoming popular. Unfortunately, users experience slow Internet connection through existing systems because paths used to route traffic are selected randomly without considering latency between nodes. In our work, we aim to design a better approach for selecting paths in anonymity systems with improved performance and without sacrificing privacy. We consider stratified topologies for our design where nodes are divided into three hops. We propose a scheme to build restricted network topologies on top of a stratified topology that maximizes throughput. First, we use Tabu Search to build latency-aware stratified restricted topologies that select paths with low latency. Then we extend this approach for heterogeneous bandwidth and propose a bandwidth scheme to build multi-link stratified restricted topologies, with bandwidth capacity equally shared by each link. Using the reduced-overhead DLP scheme for padding, we measure the anonymity of our topologies by calculating entropy on the sender probability distribution. We evaluate our system in simulator by running traces of real Tor traffic through each topology. We compare our results with several restricted topologies based on brute force and greedy approaches. We show that our proposed topologies provide 22% gain in performance with no increase in dummy traffic overhead while maintaining reasonable levels of anonymity.