Anonymity and Performance Analysis of Stream Isolation in Tor Network

Abstract

Tor is an overlay network software which enables users to communicate anonymously over the web. There have been numerous improvements to the original Tor design mainly to upgrade its performance and overall user anonymity. This paper explores the impact of a proposal to separate client based streams over the circuits, based on the metadata provided by the connection. Prior to the implementation of stream isolation, all the streams from the client end selected the circuit based on the exit node and the lifetime of the circuit. This could result in unrelated traffic being combined and let the adversary at the exit node perform probabilistic timing attacks, thereby compromising the overall anonymity of the user. Stream isolation obfuscates these attacks by creating a new circuit and isolating the streams as much as possible. We have analyzed and examined the effects of stream isolation in the Shadow and Tor Path simulator(TorPS) by creating up to ten parallel transfers from the client end which connects to several non-identical servers. These parallel transfers create distinct circuits for every request in the case of the configuration where the stream isolation is enabled. The impact on the anonymity has been investigated using TorPS, which clearly indicates a diminution in the fraction of compromised streams after isolating disparate traffic across discrete paths. The performance insights attained after the Shadow simulations show that the stream isolation when enabled in the configuration, results in an upsurge in the memory usage, increase in the latency for finishing all the downloads, and an overall rise in the network throughput. We can, therefore, conclude that this increased overall anonymity of the network comes at the cost of an increase in the network overhead.