Analyzing PeerFlow – A Bandwidth Estimation System for Untrustworthy Environments

Abstract

Tor is the most popular low-latency anonymization network comprising over 7,000 nodes run by volunteers. To balance the user traffic load over the diverse resource capabilities of these nodes, Tor guides users to choose nodes in proportion to their available bandwidth. However, self-reported bandwidth values are not trustworthy. Recently, a new bandwidth measurement system, PeerFlow, has been proposed aiming to solve the Tor bandwidth estimation problem. In this work, we introduce the first practical analysis of PeerFlow. We proposed a set of strategies for the practical realization of probation periods in PeerFlow and showed that many Tor nodes cannot recover to their normal state after one measuring period. We also demonstrated that low-bandwidth adversaries gain significantly higher bandwidth estimates exceeding the theoretically defined security boundaries of PeerFlow.