Modeling Adaptive Video Streaming Using Discrete-Time Analysis.

HTTP Adaptive Streaming (HAS) is the de-facto standard for video delivery over the Internet: video clips are split into segments and for each segment, several quality levels are provided. Based on client-centric parameters like the video buffer state or the throughput, an adaptive bitrate (ABR) algorithm is used to decide the quality of the segment to be downloaded. Besides the applied ABR algorithm, network and video characteristics, as well as the setting of buffer thresholds, the number of provided quality levels, and the choice of segment durations influence the Quality-of-Experience (QoE). Thus, the QoE is the result of the highly complex interaction of these parameters and the interdependence of these parameters has not been explored in a holistic manner yet. In this paper, a generic performance model for rate-based and buffer-based ABR algorithms is proposed. Using discrete-time analysis, the model allows computing relevant HAS metrics such as video interruptions and playback quality. To highlight the practical applicability of the model, an extensive evaluation is presented, in which the model's probabilistic results are compared with actual measurements of buffer-based and rate-based ABR algorithms. The results indicate that the model is sufficiently expressive to capture the HAS behavior across various settings, while still remaining computationally tractable.