Network Throughputs Modelling for Mobile Video Streaming Analysis

Abstract

Continuous increasing the share of video streaming in Internet traffic has raised interest in research aimed at improving bit rate selection algorithms for HTTP Adaptive Streaming (HAS). These algorithms use a different bitrate adaptation logic to ensure adaptation to the change of available network bandwidth. Measured throughput traces for different networks are usually used in the analysis of the performance of the algorithms. Although real networks are best presented by these measured traces, due to large differences in statistical properties of traces measured at different times of the day or year, as well as different routes of users in mobile networks, they are not fully suitable for systematic analysis of streaming algorithms. For this purpose, synthetic (computer-generated) traces can be more appropriate, provided that they mimic realistic traces well. In this paper, we present results of statistical modeling of throughput traces by using Nakagami distribution. Based on the parameters of Nakagami distribution estimated from the data measured in 3G and 4G networks, synthetic traces were generated. By applying the simulation framework for video streaming and using the Liu adaptation algorithm, parameters that affect the quality of the uploaded video have been compared for the cases of using synthetic and real traces. Comparison of achieved average quality levels, estimated bandwidths, video bitrates as well as number and depth of quality switches show that Nakagami distribution is a good choice for generating the synthetic traces. By changing the shape parameter m and the spread parameter of the probability density function of the Nakagami distribution, synthetic traces can be generated that correspond to a wide range of real traces.