On-Line Flows Classification of Video Streaming Applications

Delivering video streaming services over a best-effort packet network, such as the Internet, is complicated by a number of factors including time-varying bandwidth, delay, and losses. Different solutions have been proposed to optimize Live video streaming (e.g., Multicast) and to optimize VOD streaming (e.g., Cache). Implementation of these solutions at an Internet service provider (ISP) network or at Content Data Network (CDN) can use on-line classification capability of video streaming flows (or sources) into Live streaming type and VOD streaming type to allow proper optimization. Since all streaming applications use the same streaming protocols, Deep Packet Inspection (DPI) technologies are practically useless in classifying such applications. In this paper we formulate the problem of on-line video-streaming flow classification. Then, we propose and evaluate two statistical on-line streaming flow classifiers. These classifiers are based on the statistical characterization of the flow packets length. Afterward, we study a slightly different video streaming classification problem, in which we can assume that there are at least X concurrent flows from the same video source, X>1. We propose and evaluate an on-line classifier that decides whether these X flows are Live or VOD. This classifier is based on a simple observation that live streaming flows from the same source transfer the same information almost at the same time, while VOD flows from the same source have larger information offset. The classifiers performance evaluation is based on real traffic dataset. Our single flow best classifier tags 96% of the streams correctly, while our multi-flow classifier successfully tags 96.53% of streams for X=2. We also demonstrate a more complex multi-streaming comparing function that improves the success rate of our algorithm to 97.53% for X>2, but it clearly decreases the algorithm scalability. Finally, additional contribution of this paper includes statistical characterization of live video streaming traffic vs. VOD streaming traffic.