As-awareness in Tor path selection

Abstract

Tor is an anonymous communications network with thousands of router nodes worldwide. An intuition reflected in much of the literature on anonymous communications is that, as an anonymity network grows, it becomes more secure against a given observer because the observer will see less of the network. In particular, as the Tor network grows from volunteers operating relays all over the world, it becomes less and less likely for a single autonomous system (AS) to be able to observe both ends of an anonymous connection. Yet, as the network continues to grow significantly, no analysis has been done to determine if this intuition is correct. Further, modifications to Tor's path selection algorithm to help clients avoid an AS-level observer have not been proposed and analyzed.Five years ago a previous study examined the AS-level threat against client and destination addresses chosen a priori to be likely or interesting to examine. Using an ASlevel path inference algorithm with improved accuracy, more extensive Internet routing data, and, most importantly, a model of typical Tor client AS-level sources and destinations based on data gathered from the live network, we demonstrate that the threat of a single AS observing both ends of an anonymous Tor connection is greater than previously thought. We look at the growth of the Tor network over the past five years and show that its explosive growth has had only a small impact on the network's robustness against an AS-level attacker. Finally, we propose and evaluate the effectiveness of some simple, AS-aware path selection algorithms that avoid the computational overhead imposed by full AS-level path inference algorithms. Our results indicate that a novel heuristic we propose is more effective against an AS-level observer than other commonly proposed heuristics for improving location diversity in path selection.

Congestion-Aware Path Selection for Tor

“…Different criteria have been proposed as possible factors in the path selection algorithm, such as autonomous system awareness [11], application awareness [24], or game-theoretic criteria for maximal utility [30]. In this paper, we describe a modification to Tor's existing path selection algorithm to incorporate latency, which reduces congestion and improves load balancing.…”

CellFlood: Attacking Tor Onion Routers on the Cheap

“…Other attacks study the potential threat of a (semi) global adversary (Murdoch and Zieliński [17], Edman and Syverson [6], Chakravarty et al [3]), although this does not fit into Tor's original adversary model. Specific attacks to Tor bridges and Hidden Services have been studied by McLachlan and Hopper [15], and by Murdoch [18], respectively.…”

Onion routing circuit construction via latency graphs

“…Examples include the use of reputation-based strategies [32], opportunistic weighted network heuristics [30,33], game theory [34], and system awareness [35]. Compared to those previous efforts, whose goal mainly aim at reducing overhead via bandwidth measurements while addressing the classical threat model of Tor [4], our approach takes advantage of latency measurements, in order to best balance anonymity and performance.…”