Networking in a heterogeneous world: Optimization, measurement, and opportunities

Over the past decade, research on video streaming systems has often relied on simplified user studies that fail to capture the diversity and complexity of real-world user behavior, who contribute different economic and social values to content providers. This study seeks to address this oversight by developing and evaluating video streaming systems that aim to enhance provider revenues and take into account the dynamic behaviors of users. First of all, content providers face challenges with delivering content due to bandwidth constraints beyond their control. In this study, we introduce FlowTele, a novel system for optimizing internet traffic through remote traffic shaping, diverging from traditional Transmission Control Protocol (TCP) fairness. FlowTele, designed without the need for in-network or special client support, has shown to enhance providers' revenues by 20%--30% through strategic bandwidth reallocation in various network conditions. Additionally, the research investigates Quality of Experience (QoE) fairness among other metrics that content providers can optimize with FlowTele, offering a solution to bypass bandwidth bottlenecks while maintaining TCP-friendly operations. Secondly, to enhance video streaming quality, adaptive bitrate streaming (ABR) techniques and QoE metrics are utilized. Traditional methods often ignore user interactions like seeking or replaying, leading to a gap in understanding streaming quality. To fill this gap, we collected and analyzed 229,178 audience retention curves from YouTube and Bilibili, which we called vRetention, revealing significant variations in user engagement across different content types and regions. This data aids in refining ABR designs and QoE assessments, offering a deeper insight into user behavior for improved streaming experiences. Finally, evaluating the real-world QoE for ABR is difficult due to traditional evaluations overlooking dynamic user behaviors (e.g., seeking or abandoning videos). This research addresses this by analyzing audience retention curve data and developing mRetention, a dynamic behavior model to assess ABR performance. Findings indicate significant QoE degradation under dynamic behaviors, highlighting issues like overbuffering across network types and insufficient consideration for varied playback speeds.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Chen, Bo-Rong

Contributors dc:contributor

• Hu, Yih-Chun
• Bailey, Michael
• Godfrey, Brighten
• Mittal, Radhika

• Internet Bandwidth Management
• Content Delivery
• End-to-end User Experience
• Remote Traffic Shaping
• Co-bottleneck Detection
• Audience Retention Curve
• Youtube
• Bilibili

Statement dc:rights

• Copyright 2024 Bo-Rong Chen

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/124500

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F124500