Unveiling the End-User Viewport Resolution From Encrypted Video Traces

Video streaming is without doubt the most requested Internet service, and main source of pressure on the Internet infrastructure. At the same time, users are no longer satisfied by the Internet's best effort service, instead, they expect a seamless service of high quality from the side of the network. As result, Internet Service Providers (ISP) engineer their traffic so as to improve their end-users' experience and avoid economic losses. Content providers from their side, and to enforce customers privacy, have shifted towards end-to-end encryption (e.g., TLS/SSL). Video streaming relies on the dynamic adaptive streaming over HTTP protocol (DASH) which takes into consideration the underlying network conditions (e.g., delay, loss rate, and throughput) and the viewport capacity (e.g., screen resolution) to improve the experience of the end user in the limit of the available network resources. In this work, we propose an experimental framework able to infer fine-grained video flow information such as chunk sizes from encrypted YouTube video traces. We also present a novel technique to separate video and audio chunks from encrypted traces based on Gaussian Mixture Models (GMM). Then, we leverage our dataset to train models able to predict the class of viewport (either SD or HD) per video session with an average 92% accuracy and 85% F1-score. The prediction of the exact viewport resolution is also possible but shows a lower accuracy than the viewport class.