Traffic Classification for Mobile Video Streaming Using Dynamic Warping Network

Abstract

Traditional traffic classification methods generally sort the Internet traffic for video streaming to the same category. However, video streaming should be treated differently according to different streaming techniques for the task of QoE evaluation. Meanwhile, end-to-end encryption and different encrypted forms make traffic classification even more challenging because of insufficient distinguishable characteristics. In this work, we propose a novel Dynamic Warping Network (DWN) model that allows us to differentiate different streaming techniques based on traffic patterns. We compute soft Dynamic Time Warping (DTW) distances between the download speed series and a set of warping series, which are further fed to Multi-Layer Perceptions (MLP) for traffic classification. We also show how to train the MLP and warping series jointly using back-propagation algorithm. The proposed model outperforms the start-of-the-art MaMPF model for distinguishing Internet traffic between different streaming techniques, where the accuracy for Video on Demand (VoD) using HTTP Adaptive Streaming (HAS) and live broadcasting (LB) reaches 90.51% and 88.84% respectively.