Accurate Identification of Internet Video Traffic Using Byte Code Distribution Features.

Video traffic, the most rapidly growing traffic type in Internet, is posing a serious challenge to Internet management. Different kinds of Internet video contents, including illegal and adult contents, make it necessary to manage different video traffic using different strategies. Unfortunately, there are few research work concerning Internet video traffic type identification. In this paper, we propose a new effective feature extraction method, namely byte code distribution (BCD), for Internet video traffic type identification. The BCD method first counts the times of each byte code value (0 to 255) from a video flow, and then computes the ratio between each count and the total byte count. Such that the 256 ratios are used as the features. Comparing with traditional packet-level features, the BCD features contain more video type information, and are able to make identification more accurately. To test the performance of our proposal, we collect a set of video traffic traces containing two typical video types, romance and action. We conduct a set of comparing experiments on the collected data. The results show that the BCD method can hit extremely high identification accuracies (higher than 99%), far higher than those of the traditional packet-level feature extracting methods. The empirical studies show that the BCD method is promising for Internet video traffic identification.