Enabling Robust, Secure, and Efficient Cellular Networks with Fine-Grained Radio-Layer Analytics

Mobile devices have now become indispensable to the lives of billions of people around the world. Cellular traffic driven by the use of these devices is projected to increase 7x by 2021 from its levels in 2016, exacerbating already-severe network congestion and security concerns. In this thesis, we respond to these challenges, focusing especially on their impact on popular and nascent media applications such as video streaming, video conferencing, virtual reality, and autonomous vehicles. We show that granular radio-layer analytics provides a powerful lens into network dynamics, exposing patterns in resource allocation that can be used (1) to diagnose security, efficiency, and robustness challenges afflicting mobile networks, and (2) to develop solutions to these problems.

First, we present the design and implementation of eNBsniffer, a passive tool that enables us to characterize the congestion state of any cell tower, using only the fine-grained radio resource allocation data that it streams for all of its users, without requiring any privileged network information. We test it extensively in the field and show that it performs with an error rate not exceeding 5%.

We then demonstrate two use cases of eNBsniffer, revealing its capabilities as a diagnostic engine to identify and address security and efficiency concerns that affect mobile networks. We show that broad classes of popular mobile applications have distinct radio resource allocation signatures, design an application classifier based on this insight, and show that anyone can accurately infer the types of applications being served by any given cell tower using eNBsniffer. We explore the privacy implications of this finding and propose solutions to conceal application identity.

We also present BurstTracker, a tool that mobile application developers can use to pinpoint the root causes of poor application performance on cellular networks. We discuss the design of BurstTracker, which was inspired by our observations with eNBsniffer, and show how it can be used to determine whether the performance of an application is limited by congestion at the cell tower (which an application developer cannot control) or by another network inefficiency. We then demonstrate how BurstTracker is used in practice; we apply it to study mobile video streaming, explain why video performs poorly on cellular networks, and develop a solution that improves video quality on congested networks by 35%.

We conclude by discussing the broader implications of this work, highlighting how the fine-grained radio-layer analytics provided by eNBsniffer can be used to answer a host of societally-relevant questions.