PredicTor: Predictive Congestion Control for the Tor Network

Routing traffic through the Tor network is a complex task due to relaying traffic over multiple nodes of the network. This architecture, while providing the required anonymity, poses significant drawbacks in terms of latency and fairness. The main issue are poorly scheduled transmissions of data. With PredicTor we seek to solve this problem by applying distributed model predictive control (MPC) to the Tor network. In this work we formulate a novel fairness based optimization problem and present a predictive control algorithm based on this formulation. We investigate the proposed method in a simulation study based on the high-fidelity network simulation tool ns-tor (a variant of ns-3). PredicTor significantly outperforms traditional approaches in terms of fairness and latency.

The full paper can be found on arXiv.

Back