Traffic classification on mobile core network considering regularity of background traffic

Abstract

Recent widespread use of smartphones and rich multimedia contents has brought a considerable increase in mobile traffic. Therefore, the characteristics of smartphone traffic need to be considered when designing mobile core networks. Smartphone traffic is categorized by whether it is generated from user interaction (foreground (FG) traffic) or not (background (BG) traffic), and such traffic types can be managed differently to efficiently maintain the communication quality in the case of network congestion. However, it is difficult to distinguish such traffic types and their characteristics solely by IP addresses/port numbers. In addition, the increase in HTTPS traffic makes application-level packet inspection difficult. In this paper, we propose a traffic classification method on a mobile core network. The proposed method captures packets on the mobile core network, constructs TCP flows, and labels each flow by considering the regularity of BG traffic and the randomness of FG traffic using a Support Vector Machine (SVM) classifier with selected feature indexes obtained from the TCP/IP layer. Also, we attempt to restrict the indexes that can be obtained in the observation of the initial part of each TCP flow in order to reduce the calculation cost. The proposed method is evaluated in terms of the classification accuracy using experimental data through the Wi-Fi connections of smartphones. The full-index classifier using 40 indexes classifies FG and BG traffic with 97.2% accuracy and the shortcut-index classifier using restricted 36 indexes also indicates 94.4% accuracy.