[34892] in North American Network Operators' Group
Re: Network for Sale
daemon@ATHENA.MIT.EDU (Paul A Vixie)
Wed Feb 21 17:07:00 2001
Message-Id: <200102212118.NAA50618@redpaul.mfnx.net>
To: nanog@nanog.org
In-Reply-To: Message from "Steven M. Bellovin" <smb@research.att.com>
of "Wed, 21 Feb 2001 15:43:18 EST." <20010221204318.9EE9435C42@berkshire.research.att.com>
Date: Wed, 21 Feb 2001 13:18:20 -0800
From: Paul A Vixie <vixie@mfnx.net>
Errors-To: owner-nanog-outgoing@merit.edu
> >no, that is not the problem. oh i admit that ping time jitter is ~random.
> >but even if it weren't, RTT doesn't drive performance, (bw*delay)-loss does.
>
> And how does "delay" differ from RTT, except for the obvious constant
> factor?
my point wasn't that rtt didn't matter but that bandwidth and loss also matter.
so does filesize. for a 2MB GIF i'd rather have a 900ms * 622Mb/s (satellite?)
link. for a 2K frame i'd rather have a 60ms * 56K link. using RTT by itself
as a performance predictor is just silly. same for aspath length. the only
way to know if performance will differ between server/proxy A vs B when talking
to client C is to serve from both and see what the tcp window does and what the
total bytes/sec are (if the file is long enough for those to matter.) the data
then decays rapidly since congestion and routing will change after measurement.
