Traffic Morphing: An Efficient Defense Against Statistical Traffic Analysis.

摘要: Recent work has shown that properties of network traffic remain observable after encryption, namely packet sizes and timing, can reveal surprising information about the traffic’s contents (e.g., language a VoIP call [29], passwords in secure shell logins [20], or even web browsing habits [21, 14]). While there are some legitimate uses for encrypted analysis, these techniques also raise important questions privacy communications. A common tactic mitigating such threats is to pad packets uniform send at fixed timing intervals; however, this approach often inefficient. In paper, we propose novel method thwarting statistical analysis algorithms by optimally morphing one class look like another class. Through use convex optimization techniques, show how modify real-time reduce accuracy variety classifiers while incurring much less overhead than padding. Our evaluation technique against two published [29] [14] shows works well on wide range data—in cases, simultaneously providing better lower naive