| （英） |
Recently, Internet video traffic has been rapidly increasing due to the huge demands for video streaming services. Since P2PTV applications exchange video data among peers, server load can be reduced as compared to a client/server method. As both the number of peers and the throughput vary with respect to each content, P2PTV traffic is difficult to manage and control. A method to classify and model P2PTV traffic by focusing on the number of peers and throughput has been studied. However, the classification criteria in this study are ambiguous because the authors tried to subjectively classify P2PTV traffic. Clustering, one of machine learning methods, can classify a large amount of data into some categories by calculating the similarity based on the characteristic values of input data. We thus propose a P2PTV traffic classification method using machine learning. We analyzed the characteristics of P2PTV traffic that was obtained when we watched video contents,and then examined the proposed method. In addition, we created learning data from a large amount of P2PTV contents, and classified traffic by the proposed clustering method. The classification results show that over 400 P2PTV traffic data sets can be categorized into three clusters. |