Examination of QUIC-based Website Fingerprinting

Abstract

Recently, there has been a considerable amount of interest in extending QUIC for new capabilities. Since QUIC is governing most of the internet traffic, questions arise about its security. In this regard, the paper aims to understand the procedure of website fingerprinting in the case of QUIC websites as compared to TCP websites. It aims to provide a comparison of the effectiveness of the earlier used ML methods on the TCP traffic and QUIC traffic. Through this paper, a thorough comparative study of various machine learning models has been incorporated to identify the website to which a packet belongs to. In this paper, Wireshark has been used to capture packets from 50 different websites that use QUIC using a Perl script. The paper presents the results of training the data on a combination QUIC and TCP packets using the commonly used machine learning models p-FP, Var-CNN, CUMUL, Wang-KNN and k-FP. Results show that Random forest classification gives the best accuracy whereas k-FP gives the least. It can also be noted that the overall efficiencies of all the algorithms are much lesser on the combination dataset as compared to datasets with only TCP traces. This can be attributed to the fact that these models have been designed for TCP data, and are unfamiliar with QUIC traffic specifications.