Consumer video dominates Internet traffic and shows no signs of abating. In addition to being a popular source of entertainment, Internet video and its associated Quality of Experience (QoE) directly affects citizen engagement with society, from news that informs citizens of current events and issues to lifelong skill development that provides economic opportunity. Content providers can monitor end-user QoE on their apps and use these measurements to optimize the streaming service design. It is, however, challenging for Internet Service Providers (ISPs) to obtain end-user video QoE information as they lack access to streaming apps on user devices, the device itself, or the servers. The goal of this project is to develop scientific approaches that enable end-user ISPs to use network measurements to infer video QoE. Through collaboration with two real-world network operators, a Mobile Network Operator network and the Georgia Tech campus network, the project will enable large scale inference of video QoE across many different video services, end-user devices, and network data types.
Most video services deploy HTTP Adaptive Streaming (HAS) techniques that require no special network provisioning or resource reservations, and instead dynamically probe bandwidth availability and adjust video quality up or down based on bandwidth fluctuations. For such services, video service quality is practically and usefully estimated through video QoE metrics that are generally believed to correlate with user satisfaction, such as average bandwidth and number of video stalls. ISPs, where the end-user attaches, seek an in-depth understanding of the relationship between end-user QoE and ISP performance. This work will (1) develop QoE inference techniques for a variety of available network measurement data from different network types; (2) explore the use of session modeling as an approach to QoE inference; (3) develop a novel machine learning approach to inference that is informed by insights from session-based modeling; (4) investigate scalability techniques that tradeoff accuracy of inference with real-world feasibility; and (5) develop a framework for continuous calibration and training. This project will enable large scale inference of video QoE, leverage and develop partnerships between industry and academia, and will produce tools and data available to researchers and educators.
The project web page is www.cc.gatech.edu/~ammar/VIQI.html. Georgia Tech Smartech (https://smartech.gatech.edu) will be used as a long term repository. This will be linked through the project web page above. The goal is to preserve the project data for at least 5 years beyond the end of project.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
