Improving Mobile Internet Performance with Cross-Device Network Transport and Cross-Layer Application Adaptation

Abstract

The mobile Internet is becoming increasingly complex with a wide diversity of end systems (including wearables and automobiles), the co-presence of multiple devices with collaboration potential, and the growth of user-generated application traffic fueled by improved mobile sensors and wireless access. As the Internet evolves along with these trends, the increased complexity of different components and protocol layers makes it more challenging to achieve high network utilization and meet the diverse QoE requirements for mobile applications. As a result, despite the richness of various network resources, the performance of today’s mobile applications still falls behind expectations. To address this challenge, in this dissertation, I demonstrate that with a better understanding of the various components and different protocol layers of the increasingly complex mobile Internet, we can identify unique performance problems and leverage such knowledge to develop network transport protocols with cross-device awareness and application adaptation strategies with cross-layer considerations for better mobile app performance.

Specifically, to understand the interaction between multiple mobile devices and its impact on end-to-end performance, we conduct an empirical study on wearable networking, where wearables oftentimes rely on their paired smartphones for Internet access. Based on our measurement findings, we develop cross-device network transport management solutions for improving wearable networking performance. To better support multi-device collaboration in a more general setting, we develop MPBond, a distributed multipath transport system for efficient network-level collaboration among personal mobile devices, with cross-device connection management and packet scheduling. To explore opportunities on the mobile application design, we build Livelyzer, a generalized measurement tool for characterizing commercial live video streaming upstream ingest performance under mobile networks. Based on our measurements, we identify deficiencies in broadcasting app rate adaptation, and propose network-aware adaptation strategies to improve the same. We also investigate another emerging mobile application, vehicular sensing for autonomous driving, where we build Harbor, a cross-layer system architecture for collaborative vehicular sensor data sharing, with adaptive usage of V2V and V2I network resources.