MobInsight

Abstract

It is well-known that the performance of Web-browsing as well as mobile applications (or apps) suffers on today's cellular networks. In this work, we perform a systematic measurement study of more than 50 popular apps and 2 cellular networks, and discover that while cellular networks have predictable latency, it is the path between exit points of cellular networks (e.g., GGSN) and cloud-servers that degrades apps performance. High latency and unpredictability over this path affects browsing and activity completion times of apps, worsening the performance by several magnitudes. Furthermore, we find that as the number of apps on mobile devices increases, cellular networks in turn suffer due to large number of active connections, primarily used for push notifications, experiencing heavy signaling overhead in the network. Towards accelerating the performance of apps and improving their operational efficiency, we envision an easy to deploy operator-managed platform, and study two architectural optimizations that sit at vantage points inside cellular networks: virtual app-server (vApp) and network-assisted, virtual push-notification server (vPNS). vApps improve apps' browsing experience while vPNSs take the burden of carrying periodic message off cellular networks. Using trace-driven simulations, we find that vApps can improve activity completion times by more than 3-fold, whereas vPNS can reduce the signaling load by a factor of 6 in cellular networks and reduce energy consumption by a factor of 2 on mobile devices.