[38372] in North American Network Operators' Group
Re: 95th Percentile != Lame
daemon@ATHENA.MIT.EDU (mike harrison)
Mon Jun 4 08:18:06 2001
Date: Mon, 4 Jun 2001 08:17:21 -0400 (EDT)
From: mike harrison <meuon@highertech.net>
To: "Richard A. Steenbergen" <ras@e-gerbil.net>
Cc: David Klindt <dklindt@cobra.ordata.com>,
James Thomason <james@divide.org>,
"nanog@merit.edu" <nanog@merit.edu>
In-Reply-To: <Pine.BSF.4.21.0106032244240.29677-100000@overlord.e-gerbil.net>
Message-ID: <Pine.LNX.4.10.10106040810470.1109-100000@home.highertech.net>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Errors-To: owner-nanog-outgoing@merit.edu
> But peak vs non-peak has little to do with 95th percentile. Assuming that
> every day's traffic patterns are the same (which is rarely true, there is
My nybble's worth:
95th percentile is usually based on a 5 minute average, and when you
have a lot of upstream bandwidth you can move a lot of data quickly and
it barely shows in a graph/monitoring system doing 'mrtg' style 5 minute
samples.
It's better than nothing, and identifies 'hogs' quickly.
Because of the large amount of 'business hours' users we have, I've
made some special deals to some 'off-hour' bandwidth hogs because of the
available and unused bandwidth during those hours. --Mike--
