摘要: Webpage fingerprinting methods infer the webpages visited in a traffic trace and are serious threats to privacy of web users. Prior work evaluates webpage using samples from single client does not consider diversity factor---webpages can be different browsers, operating systems devices. In this paper, we study impact on HTTPS fingerprinting. First, evaluate 5 prominent 19 clients. We show that best performing overfit patterns do generalize when they evaluated (even if clients use same browser system only differ device). Then, investigate find differences HTTP messages generated, servers communicated implementation HTTP/2 across Finally, robustness increased by training them diverse set This informs community towards realistic threat model for presents an analysis modern traffic.