Foolproofing network communication by k-diverse padding
It has been shown recently that a large majority of online web-based applications are prone to privacy attacks, causing serious breach in the privacy of any online activity, due to the popular application features that generate network traffic patterns. Features like auto-completion and auto-suggestion inadvertently reveal underlying user actions with unique packet bursts that, despite being fully encrypted, enable an eavesdropper to determine all user activity on a web-application. Well-known techniques for achieving data privacy, such as k-anonymity and l-diversity, can be adapted in this context of web applications to achieve desired levels of privacy by padding packets with dummy bytes, aimed at obfuscating any eavesdropper by blending different user actions with each other. In this work, we achieve a high level of privacy by a novel technique that blends publicly observable network bursts of carefully chosen probabilistic portions of user actions. This technique in fact results in the maximum possible level of l-diversity, i.e. k-diversity, and in that respect is a significant improvement over all existing techniques.
Thesis (M.S.)-- Wichita State University, College of Engineering, Dept. of Electrical Engineering and Computer Science