Towards reduced latency in adaptive live streaming

Abstract:

Adaptive video streaming is today's predominant video delivery model in the Internet. However, existing bitrate adaptation schemes exhibit a latency of several tens of seconds. This latency is mainly determined by the buffering delay, which aims at absorbing the instantaneous mismatches between the network throughput and the video bitrate. A lot of research focuses on the adaptation algorithms, few researchers, however, address the achievable lower bound of the latency. Our paper fills this gap by deriving an approximation of the lower bound for the buffering delay. In this study, we analyze the main contributors to the latency and develop a mathematical model for the process of adaptive live streaming. Then, we present the required buffering delay given characteristics of a network (e. g., throughput and delay). This provides us with guidelines to determine a reasonable low latency for streaming systems. We validate the approximation under a trace-driven simulation of a mobile network in the context of video streaming. Results show that our approximation approaches to the theoretical minimum on average.