|
|
Characterization of User Access to Streaming Media Files
|
Authors
|
Jussara M. Almeida <jussara@cs.wisc.edu>
Jeffrey Krueger <jkrueger@cs.wisc.edu>
Mary K. Vernon <vernon@cs.wisc.edu>
Computer Sciences Department, University of Wisconsin-Madison
|
Abstract
|
This paper presents an analysis of server log data for two real-time
streaming media servers that deliver education content at two different
major U.S. Universities. Our analysis reveals some interesting observations
regarding file access characteristics on these servers. For example, during
periods of stable request arrival rate, the client session arrival process
is approximately Poisson, but the inter-request times for interactive requests
are better approximated by a Pareto distribution. The distribution of file
popularity are well modeled by the concatenation of two Zipf-like distributions.
Furthermore, a significant fraction of the files, or file segments, that are
requested in a given hour but not in the previous hour, are not requested
again for several hours, which motivates the need to reevaluate the traditional
cache-on-first-miss strategy for streaming media files. Finally, high client
interactivity in these environments leads to large numbers of partial video
file accesses. Nevertheless, the requirements on remote server and network
bandwidth can be reduced by 40 to 60% by using multicast delivery.
|
|