Mobile data traffic analysis: How do you prefer watching videos?

In this paper, we analyze an anonymized packet trace of mobile handset data traffic from a major European mobile Internet Service Provider (ISP). Our goal is to understand the effect of the differences between this traffic and the traffic in wired residential networks. This knowledge can be useful for adapting the network to specific mobile usage. We show that, compared to wired users, mobile users exchange data with fewer servers. We then evaluate the performance at flow level, and study the influence of the access type (2G or 3G). In particular, we quantify the improvement provided by a 3G access in terms of delays and throughput. To figure out the user experience, we focus on a popular usage: media streaming. We measure performance metrics for the two main content diffusion techniques: ISP-managed live TV and progressive download (PDL, e.g. YouTube or Deezer). We observe that live TV streams only suffer from moderate packet losses thanks to conservative encoding bitrates chosen according to the radio access type. As for progressive download, while audio streams generally cope well with mobile networks, only 3G users experience smooth video playback. Through the analysis of session durations, we determine the effect of the Radio Access Type on the user experience.We notably show that users react to impaired network conditions by aborting their streaming sessions. We also highlight usage differences between live TV and PDL.