Dive into streaming: efficient identification of encrypted dynamic DASH video traffic

Numerous online videos are transmitted over the Internet via encrypted network traffic. Video traffic identification aims to identify the encrypted traffic of specific videos, such as harmful videos that promote violence or incite suicide, which is of great significance to public safety and network regulation. However, the latest improvements in multimedia streaming have rendered existing identification methods ineffective or impacted. Besides, the complexity of real-world video traffic challenges the capabilities of existing methods. In this work, we present the first in-depth analysis of multimedia streaming that incorporates a segment-combined mechanism, revealing the traits of its video traffic and combination patterns. Then, we propose Evi, an efficient method for dynamic video traffic identification. Evi uses combination patterns learned from the training set to generate video fingerprints from multimedia segments. During identification, Evi uses both video traffic and fingerprints to build chunk transition graphs, achieving efficient identification by calculating the multi-source shortest distances on the graphs. Evaluations show that Evi only requires about 35 s of video traffic to achieve an average accuracy of 0.975 in the open-world experiments. In addition, Evi is robust to video traffic from various streaming scenarios, such as poor network conditions.