Traffic Aware Based Tail Optimization of Browsing Applications for Energy Saving

Abstract

It is challenging to save energy in smartphones by shortening simply tail time for user real-time interactive applications. Because users’ behavior is stochastic and random so that it is difficult to learn network data traffic characteristic of transmission. In this paper, we propose a novel scheme TATO, which effectively forecast long idle time between active network activities in advance to reduce unnecessary extra tail time. The core idea is to learn recent interval pattern of traffic transfer by SVM model to predict next interval time whether exceed predefined threshold thus to adjust radio interface from high power state to idle state by leveraging Fast Dormancy mechanism. Traffic pattern may be different from user to user but we design temporal correlation features to train SVM model to achieve predict accuracy up to 90 % that it can save to 70 % optimal saving energy on average without much influence on user experience.