Website Fingerprinting at Internet Scale

Abstract

Abstract-The website fingerprinting attack aims to identify the content (i.e., a webpage accessed by a client) of encrypted and anonymized connections by observing patterns of data flows such as packet size and direction. This attack can be performed by a local passive eavesdropper -one of the weakest adversaries in the attacker model of anonymization networks such as Tor.In this paper, we present a novel website fingerprinting attack. Based on a simple and comprehensible idea, our approach outperforms all state-of-the-art methods in terms of classification accuracy while being computationally dramatically more efficient. In order to evaluate the severity of the website fingerprinting attack in reality, we collected the most representative dataset that has ever been built, where we avoid simplified assumptions made in the related work regarding selection and type of webpages and the size of the universe. Using this data, we explore the practical limits of website fingerprinting at Internet scale. Although our novel approach is by orders of magnitude computationally more efficient and superior in terms of detection accuracy, for the first time we show that no existing method -including our own -scales when applied in realistic settings. With our analysis, we explore neglected aspects of the attack and investigate the realistic probability of success for different strategies a real-world adversary may follow.

Website Fingerprinting Defenses at the Application Layer

“…The latest attacks against Tor achieve more than 90% accuracy in a closed-world of websites, where the attacker is assumed to have samples for all the websites a target user may visit [5,11,20,30,31]. This assumption is unrealistically advantageous for the attacker [14] and a recent study has shown that the attack does not scale to large open-worlds [20].…”

“…This assumption is unrealistically advantageous for the attacker [14] and a recent study has shown that the attack does not scale to large open-worlds [20]. However, the .onion space is significantly smaller than the Web and may be feasible for an adversary to train on a substantial fraction of all .onion websites.…”

“…CUMUL depends on packet lengths and so some defenses that only operate on packet time information cannot be applied. use CUMUL [20], k-FP [11] k-NN [30] for evaluation 12 .…”

Personalized Pseudonyms for Servers in the Cloud

“…Also outside our scope are connection-level features that can divulge indications of the server involved in the connection, such as have been used in TCP fingerprinting (e.g., [17]), TLS fingerprinting (e.g., [26]), or website fingerprinting (e.g., [13,31,39]). Such features include the number of servers to which the client connects, the timing connections to relative to one another, connection volume patterns, etc.…”

“…While we have made initial steps to support per-connection trafficanalysis defense (Appendix A), that defense does not immediately address traffic analysis based on aggregates of connections, e.g., the number of servers to which connections are made or the relative timings of these connections. Defending against this type of attack remains a very active area of research independent of our proposal (e.g., [13,14,31,39]). …”