Improving Mobile Network Performance Through Measurement-driven System Design Approaches

Abstract

Mobile networks are complex, dynamic, and often perform poorly. Many factors affect network performance and energy consumption: examples include highly varying network latencies and loss rates, diurnal user movement patterns in cellular networks that impact network congestion, and how radio energy states interacts with application traffic. Because mobile devices experience uniquely dynamic and complex network conditions and resource tradeoffs, incorporating ongoing, continuous measurements of network performance, resource usage and user and app behavior into mobile systems is essential in addressing the pervasive performance problems in these systems.

This dissertation examines five different approaches to this problem. First, we discuss three measurement studies which help us understand mobile systems and how to improve them. The first examines how RRC state performance impacts network performance in the wild and argues carriers should measure RRC state performance from the user's perspective when managing their networks. The second looks at trends in applications' background network energy consumption, and shows that more systematic approaches are needed to manage app behavior. The third examines how Server Push, a new feature of HTTP/2, can in certain cases improve mobile performance, but shows that it is necessary to use measurements to determine if Server Push will be helpful or harmful. Two other projects show how measurements can be incorporated directly into systems that predict and manage network traffic. One project examines how a carrier can support prefetching over time spans of hours by predicting the network loads a user will see in the future and scheduling highly delay-tolerant traffic accordingly. The other examines how the network requests of mobile apps can be predicted, a first step towards an automated and general app prefetching system. Overall, measurements of network performance and app and user behavior are powerful tools in building better mobile systems.