TBP: Tree structure Burst-sequence Padding Defense Against Website Fingerprinting

Website Fingerprinting (WF) attacks have been claimed to seriously threat users’ online privacy. Such surveillance is hard to be avoided even for those browsing through Tor. Therefore, a series of WF defenses have been proposed to alleviate this threat, however, those methods usually require high computation or additional infrastructure, and with increased bandwidth overhead. In this paper, we propose a novel tree structured perspective to describe the browsing process of a website, i.e., the browser usually requests website resources (with order, nested and citation correlations, etc.) through multiple threads, hence, specific parallel and causal relationships among website resources can be reflected from the tree structured perspective. Based on that, we therefore design an efficient low cost defense called Tree structure Burst-sequence Padding (TBP). TBP adds dummy burst sequences to mimic fake website resources to traffic traces, leading to effective destruction of the original tree structure. Specifically, by using multiple random distributions to determine the insertion timetable of dummy burst sequences, TBP increases the randomness of intra-class traces and blurs the boundary of inter-class traces. Extensive experimental results show that TBP is able to reduce the attakcer’s true positive rate by 60% with 55% data overhead, saving 44% data overhead compared to the state-of-art lightweight defense FRONT. Compared with FRONT, TBP can reduce the true positive rate of the attacks, including DF, Tik-Tok, Var-CNN, by about 15%.