[36797] in North American Network Operators' Group
Re: What does 95th %tile mean?
daemon@ATHENA.MIT.EDU (Alex Pilosov)
Fri Apr 20 09:53:49 2001
Date: Fri, 20 Apr 2001 09:57:55 -0400 (EDT)
From: Alex Pilosov <alex@pilosoft.com>
To: Toby_Williams@enron.net
Cc: nanog@merit.edu
In-Reply-To: <88256A34.00334E65.00@ecmta1.enron.net>
Message-ID: <Pine.BSO.4.10.10104200953410.17529-100000@spider.pilosoft.com>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Errors-To: owner-nanog-outgoing@merit.edu
On Fri, 20 Apr 2001 Toby_Williams@enron.net wrote:
> _95th%ile is not a robust mechanism for billing_
> It neither addresses variance nor average use effectively and can be gamed
> quite easily. It's some kind of best fit system for billing buyers with "normal
> behaviour".
>
> Where in maths class did they ever say that scraping off the highest n percent
> of a data set in isolation, gives a good indication of anything? Don't you need
> to look at the mean and 90th percentile together to fairly evaulate
> distribution.
>
> I think the reason it's so popular currently is that it's easy to describe
> (hence sell) and fits normal use reasonably well, so from a "normal buyer's"
> perspective is OK. Just as long as everyone is honest.
Absolutely. Here's a scheme that, if everyone was using only 95th
percentile billing, would give you webhosting at gigabit rate while only
paying 30x minimal rate webhosting:
a) colocate hardware at 30 different datacenters at gigE speed, but billed
for 95th percentile.
b) have a DNS server that rotates records, pointing to a different
server each day.
c) bingo. Each of your providers will remove highest 95th percentile (1.5
days worth of traffic) and only bill you for minimal utilization.
-alex
