PEZD: A practical and effective zero-delay defense against website fingerprinting

Abstract

Website Fingerprinting (WF) enables an attacker to infer which website a user is visiting by analyzing the side-channel information of network traffic, posing a serious threat to anonymous systems such as Tor. To mitigate this issue, numerous defenses have been proposed to resist WF attacks. However, many defenses incur high overhead, impeding their deployment. Moreover, some defenses rely on the impractical assumption that the defender knows prior knowledge of which website the user visits. In this paper, we propose a practical and effective lightweight WF defense, named PEZD. It hinders attackers from recognizing a website’s unique patterns by adding dummy packets with diverse distributions into traffic traces. In particular, PEZD is a website-agnostic defense that does not require knowledge of the website users are visiting and incurs only moderate overhead to effectively prevent attacks. Extensive experiments demonstrate that PEZD reduces the accuracy of state-of-the-art attacks from 97 % to 19 %–38 % while introducing zero latency overhead and only 48 % bandwidth overhead. Moreover, we show that PEZD remains effective for injecting dummy packets only on the client-side, simplifying the implementation of the defense.