The contents of this legacy page are no longer maintained nor supported, and are made available only for historical purposes.

Bibliography Details

J. Merwe, S. Sen, and C. Kalmanek, "Streaming Video Traffic: Characterization and Network Impact", in WCW 2002, Aug 2002.

Streaming Video Traffic: Characterization and Network Impact
Authors: J. Merwe
S. Sen
C. Kalmanek
Published: WCW, 2002
Entry Dates: 2009-02-11
Abstract: The emergence of the Internet as a pervasive communication medium, and the widespread availability of digital video technology have led to the rise of several networked streaming media applications such as live video broadcasts, distance education and corporate telecasts. This paper studies the traffic associated with two major categories of streaming content-on-demand streaming of pre-recorded content and live broadcasting. Using streaming logs from a commercial service, we analyze the traffic along a number of dimensions such as session characterization, object popularity, protocol choice and network load. Among our findings, (i) high bandwidth encodings account for about twice as many requests as low bandwidth ones, and make up about 94% of the traffic, (ii) Windows Media streams account for more than 75% of all requests, when the content is available in both Windows and Real formats, (iii) TCP based transport protocols dominate over UDP being used in about 70% of all bytes transfered (iv) Object popularities exhibit substantial skew with a few objects accounting for most of the load, (v) A small percentage of IP addresses (or routing prefixes or origin autonomous systems (ASes)) account for most of the traffic demand across a range of performance metrics. This last behavior suggests that substantial bandwidth efficiency can be realized with a distribution infrastructure comprised of a relatively small number of replicas, placed close to the heavy-hitter ASes. We also found very high variability in terms of the traffic volume with an order of magnitude or more increase in the offered load over tens of minutes, suggesting the potential benefit of a shared infrastructure that can exploit statistical multiplexing.
  • datasets: logs from a commerical streaming service;
  • analyze traffic along a numbe of dimensions such as session characterization, object popularity, protocol choice and network load;