The Pareto or Truncated Pareto Distribution? Measurement-Based Modeling of Session Traffic for Wi-Fi Wireless Internet Access, Edward Chlebus, Gautam Divgi in Proc. of WCNC'07, Hong-Kong, March 2007.
This paper was pointed out to me by its first author at the conference, and I eventually got to read it. It is a pretty simple analysis of data gathered from an Australian hot-spot operator, and it finds the session length distribution for session connection at a wifi hotspot which ends connection over 5 hours long. Thus the truncated Pareto distribution. It is pretty basic statistical analysis: discovering a distribution that fits by observation of the data, then fitting the model parameters to get a good match.
This could be useful when actually simulating wifi sessions. I am not sure there is a significant difference in the behavior of wifi users on an unlimited plan wrt to the other internet users (say, DSL), so those models should be similar. I guess the truncation is the novelty introduced by the data here, as I recall that internet traffic follows a Pareto distribution or some power law. I have to get back to On the self-similar nature of Ethernet traffic. Interestingly enough, there is no mention of the difference between wired and wifi users here.