An Empirical Study of Chunk Size Variation in DASH System

Abstract

Most video players use Dynamic Adaptive Streaming over HTTP (DASH) to ensure high Quality of Experience (QoE) in varying network conditions. To improve the encoding efficiency, streaming providers use Variable BitRate (VBR) encoding for video contents, where the video chunks with the same nominal bitrate (video quality) have different instantaneous bitrates and chunk sizes. However, the chunk size variation is assumed to be trivial compared with the network bandwidth variation, and existing bitrate adaptation algorithms simply use the nominal information to make bitrate decisions. In this paper, we present an empirical study of chunk size variation in DASH to explore whether the algorithm can ignore this stochastic factor. We measure the characteristics of chunk size variation based on a rich set of videos and analyze its impact on QoE by trace-driven experiments. Analysis results show that all the buffer-aware algorithms suffer from degraded QoE if they do not consider the chunk size variation. To eliminate the negative impact, we suggest that the bitrate adaptation algorithm should rely on exact chunk size to adjust bitrate instead of the nominal bitrate. To confirm our insights and suggestion, a preliminary solution is proposed and validated based on a real video player and realistic bandwidth traces.