A system testbed for modeling encrypted video-streaming service performance indicators based on TCP/IP metrics

Abstract

For cellular operators, estimating the end-user experience from network measurements is a challenging task. For video-streaming service, several analytical models have been proposed to estimate user opinion from buffering metrics. However, there remains the problem of estimating these buffering metrics from the limited set of measurements available on a per-connection basis for encrypted video services. In this paper, a system testbed is presented for automatically constructing a simple, albeit accurate, Quality-of-Experience (QoE) model for encrypted video-streaming services in a wireless network. The testbed consists of a terminal agent, a network-level emulator, and Probe software, which are used to compare end-user and network-level measurements. For illustration purposes, the testbed is used to derive the formulas to compute video performance metrics from TCP/IP metrics for encrypted YouTube traffic in a Wi-Fi network. The resulting formulas, which would be the core of a video-streaming QoE model, are also applicable to cellular networks, as the test campaign fully covers typical mobile network conditions and the formulas are partly validated in a real LTE network.