On Sun, 16 Feb 2003, Jonas Stenling wrote:
> I would like to continue our discussion about pricing models. Bill made
> reference to some papers stating that 95th percentile billing is flawed.
> It would be great if he could post a URL to those papers!
I would be interested in that as well.
If one defines 95th percentile billing as 5 minute averages of traffic,
sorted by amplitude on a monthly basis, and then billed at the 1/20th
highest traffic measurement, I do not think 95th percentile billing is
flawed, quite the opposite, it's a good model. I like it because it
enables the customer to burst in shorter intervals which is good for
spikey internet traffic (which is often the case).
For "safety" the customer should be shaped/ratelimited at a certain
predecided limit to put a maximum on the amount of money which can be
billed per month.
For a web server customer a good scheme might be:
95th percentile billing
minimum billing for 10 megabit/month
shaped at 30 megabit/s
a certain amount of money per megabit measured.
This enables the web server customer to grow traffic as his server usage
increases, hopefully with a business model that actually increases revenue
as traffic grows, which means that they can pay their bandwidth bill. It
stops the bandwidth bill to become too horrific in case something goes
wrong (very popular file put up by mistake for instance).
Mikael Abrahamsson email: [log in to unmask]