Modeling, identifying, and simulating Dynamic Adaptive Streaming over HTTP

As HTTP-based streaming video applications have grown to become a major source of Internet traffic, and as the new ISO standard Dynamic Adaptive Streaming over HTTP (DASH) gains industry acceptance, researchers need the ability to both study real-world viewing data and simulate realistic DASH streams. The first effort is complicated by the fact that network researchers are often restricted to anonymized, header-only traces. The second effort is difficult since researchers are currently limited to three undesirable options for generating DASH traffic: (1) encode and store movies, which is both labor-and storage-intensive, (2) parameterize DASH traffic, which is open to criticism, or (3) stream movies from a service such as Netflix, which is prohibitively expensive if a researcher wants to simulate many simultaneous viewers. In this paper we present our work towards developing a model for DASH traffic and show how the model can be applied to identify DASH streams from anonymized, header-only traces using a combination of Hadoop and Hive. We then describe how the insight gained from our work will be used in a simulator that can recreate the DASH payloads of real movies using a format that requires only a few kilobytes of storage per movie.