Improving ABR Video Streaming Design with Systematic QoE Measurement and Cross Layer Analysis

Abstract

Adaptive Bitrate streaming (ABR) has been widely adopted by mobile video services to deliver satisfying Quality of Experience (QoE) over cellular network with time-varying bandwidth conditions. To build an ABR service, a wide range of critical components spanning different entities need to be determined. It is challenging to achieve designs with good QoE properties, as the streaming performance depends on complex interactions among the various factors. To make it more complex, many design decisions also involve tradeoffs among different QoE metrics.

To address this challenge, in this dissertation, we build four systems to provide systematic support for video QoE measurements and cross-layer analysis. First, we build a general black-box measurement platform based on standard ABR protocols and common UI designs. It analyzes HTTP information in the network traffic and correlates UI events of mobile video apps to reveal ABR design and identify QoE issues. Second, to address the challenge brought by increasingly adopted encryption protocols such HTTPS and QUIC, we develop a technique called CSI to infer ABR video adaptation behavior based on packet size and timing information still available in the encrypted traffic. Third, we explore a conceptually very different approach to QoE measurement --- utilizing the on-device recording capability to record the video displayed on the mobile device screen and measuring delivered QoE from this recording. We design a novel system VideoEye to conduct such screen-recording-based QoE analysis. Lastly, to understand the interaction of existing video streaming system design with the new transport protocol QUIC, we build a platform WIQ to perform what-if analysis and measure the video QoE impact of QUIC without the need of modifying the server or client implementation. Leveraging these systems, we perform measurements on popular streaming services, understand the QoE implications of various ABR design, identify a wide range of QoE issues and develop best practices.