Abstract

The Tor network gives anonymity to Internet users by relaying their traffic through the world over a variety of routers. This incurs undesirable latency, and we explore where this latency occurs. Experiments discount transport latency and computational latency to determine there is a substantial component that is caused by delay. We determine that congestion control is causing the delay.

Tor multiplexes multiple streams of data over a single TCP connection. This is not the proper use of TCP , and as such results in the improper application of congestion control. We illustrate an example of this occurrence on a Tor node in the wild and also illustrate how packet dropping and reordering cause interference between the multiplexed streams.

Our solution is to use a TCP -over-DTLS transport between routers, and give each stream of data its own TCP connection. We give our design for our proposal, and show experiments evidence to illustrate that our proposal has in fact resolved the multiplexing issues discovered in our system performance analysis. The future work gives a number of steps towards optimizing and improving our work.