Unit Traffic Classification and Analysis on P2P Video Delivery Using Machine Learning

Abstract

This paper proposes a new traffic classification method for peer-to-peer-based video streaming (P2PTV) by using machine learning. Since P2PTV is a promising technology for video delivery, it is important to comprehend its traffic characteristics. Our previous studies classified and analyzed traffic obtained in viewing P2PTV on a per content basis. However, users' participation would be dynamic, and then the traffic characteristics may change momentarily. In the proposed method, we divide per-content traffic into short-time data units and then classify them by machine learning. From the experimental results on 80 contents data, six types of unit traffic patterns appear. Moreover, we also analyze their occurrence patterns.