Towards Optimization-Based Predictive Congestion Control for the Tor Network

Abstract

Providing online anonymity to a broad range of Internet users, the Tor network today faces not only security concerns, but increasingly also performance issues. Due to its multi-hop nature, proper congestion control has been identified to be challenging in this situation. In this paper, we focus on PredicTor, a novel approach towards multi-hop congestion control based on distributed model predictive control (MPC), an advanced optimization-based control technique. We investigate PredicTor's significance for congestion control research. In particular, we carry out a simulation study to evaluate PredicTor's performance in non-trivial network scenarios. Our results indicate the great potential to push the status quo of congestion control, heavily improving achieved latency and fairness. By pointing out benefits and challenges of distributed MPC in this context, we open up a new promising research direction for congestion control