Assessing QoE of on-demand TCP video streams in real time

Real-time stream quality assessment can assist network operators, content providers, streaming servers, and ISPs in evaluating their customers' quality of experience (QoE), and can lead to better design of streaming protocols, computer networks, and content delivery systems. In previous work, we demonstrated that objective data collected from an instrumented media player application can be used to predict video QoE of UDP streams with a high degree of accuracy and without the inherent scalability and interpretation issues of subjective ratings. In this paper, we extend our previous work to consider real time QoE assessment of TCP video streams, using as little as 15 seconds worth of data. By examining two video-on-demand scenarios and a general video scenario in which content changes rapidly, we find that we can accurately assign user quality ratings between 75 and 87% of the time using objective application-layer measurements from TCP video streams, such as bandwidth and frame rate, and that various combinations of stream state measurements produce accurate ratings in this range. We discuss the implications of these findings on the design of a real time video QoE assessment system.