I am considering getting a burstable T1 from UUNET. They compute bandwidth utilization by sampling actual bandwidth usage every five minutes for the month, then discard the highest 5% and use the 95%-th highest bandwidth utilization rate as the rate for the entire month. Depending on the bandwidth utilization distribution function, this means they are overcharging by quite a bit (88% for a linear distribution function). I want to argue with them but don't have the data to prove my point.
So my question is: can anyone point me to data showing what a typical bandwidth utilization distribution function (or probability density function) -- by bandwidth, not by time -- is for a small ISP?
Thanks in advance,