Statistical Characterization of the Chunk Size Distribution in DASH

Abstract

We present a statistical characterization of the segment size distribution of video streaming services based on dynamic adaptive streaming over HTTP (DASH). We first obtain the empirical distributions from traffic captures of available data sets encoded with different quality and segment size duration. Then, we determine the distributions that provide a better fit to the empirical data. We show that Weibull and truncated logistic distributions are adequate to modeling the chunk size distributions for a wide range of video qualities and segment size duration. Results are used to develop a source model for DASH traffic, implementing a synthetic generator of DASH-like video traces.