Close to Optimum?

Abstract

HTTP Adaptive Streaming (HAS) is the de-facto standard for over-the-top (OTT) video streaming services. It allows to react to fluctuating network conditions on short time scales by adapting the video bit rate in order to avoid stalling of the video playback. With HAS the video content is split into small segments of a few seconds playtime each, which are available in different bit rates, i.e., quality level representations. Depending on the current conditions, the adaptation algorithm on the client side chooses the appropriate quality level and downloads the respective segment. This allows to avoid stalling, which is seen as the worst possible disturbance of HTTP video streaming, to the most possible extend. Nevertheless, the user perceived Quality of Experience (QoE) may be affected, namely by playing back lower qualities and by switching between different qualities. Therefore, adaptation algorithms are desired which maximize the user’sQoEfor the currently available network resources. Many downloading strategies have been proposed in literature, but a solid user-centric comparison of these mechanisms among each other and with the global optimum is missing. The major contributions of this work are as follows. A proper analysis of the influence of quality switches and played out representations on QoE is conducted by means of subjective user studies. The results suggest that, in order to optimize QoE, first, the quality level of the video stream has to be maximized and second, the number of quality switches should be minimized. Based on our findings, a QoEoptimization problem is formulated and the performance of our proposed algorithm is compared to other algorithms and to the QoE-optimal adaptation.