Framework and Implementation of Online Smartphone Traffic Classification According to Quality Sensitivity

Currently smartphones have become the most prevalent device used on the Internet and they generate a substantial portion of Internet traffic. Smartphones can run a wide variety of applications, each generating distinctive traffic. From the perspective of network operation, classification of these types of traffic in order to prioritize quality-sensitive traffic over best-effort traffic has the potential to improve network performance. Network operators generally treat web browsing, Voice over Internet Protocol (VoIP), video streaming, and other interactive traffic as quality-sensitive traffic. This quality-sensitive traffic is generally generated by user interactions on the current foreground application, hence its communication quality is likely to affect user's quality of experience (QoE) directly. On the other hand, traffic generated when the smartphone is inactive has a low potential for affecting user's QoE. As a traffic classification technique, machine learning is popular and demonstrates high accuracy; however, it is disadvantageous in that it involves gathering enough training data sets for supervised training. We propose a framework for online smartphone traffic classification that classifies traffic according to whether it is generated when the smartphone is active or inactive. A key feature of the proposed framework is automation of training data collection for machine learning. An application on the smartphone periodically estimates whether that smartphone is active or not, then it uploads the estimation results with timestamps as a source of training data sets. We also implemented a prototype of the proposed framework and assessed the feasibility of the proposal.