A QoE-based Governor for Web Browsing on Heterogeneous Mobile Systems

Mobile browsers are one of the most popular applications on mobile devices. However, as the requirements for low latency and the increasingly complex websites, the modern browsers become energy hungry to satisfy the user experiences. Previous studies have shown that the webpage rendering contributes the most to the overall overhead. In this paper, we propose a predictive model to schedule the rendering work to achieve the purpose of maximizing the user quality-of-experience (QoE). Both of the users' delay-tolerance and energy consumption are considered in our QoE evaluation. To this end, we characterize the webpage workload and leverage a machine learning technique to find the correlation between the webpage workload and the QoE. Finally, we build a governor to predict which processor configuration can be used to execute the web rendering engine at runtime, and then schedule the rendering process based on the predicted results. Our approach are useful for not only web browsers but also a large number of mobile apps which are underpinned by web rendering techniques. We evaluate the performance of our approach on a representative mobile heterogeneous platform by using the hottest 1000 webpages, the results show high prediction accuracy for the predicted model of 85.3%. Overall, our approach delivers an average of 47.3% higher QoE than the mobile system default governor, Interactive, with an average energy reduction of 38.5% and 1.12x performance speedup.