Subject: Re: freebsd em driver
To: Jason R Thorpe <firstname.lastname@example.org>
From: Jonathan Stone <jonathan@DSG.Stanford.EDU>
Date: 06/17/2002 14:43:07
In message <20020617141152.S3049@dr-evil.shagadelic.org>Jason R Thorpe writes
>On Mon, Jun 17, 2002 at 01:58:03PM -0700, Jonathan Stone wrote:
>The "boards" in "in most boards" refers to "motherboards".
Once more, then: Motherboards with 32/33 PCI are not a sensible test
environment for gigabit ethernet. Taking performance numbers limited
by such a 32/33 PCI host environment, and treating those numbers as a
suitable performance target, is a *very* bad idea.
Or to put it another way, s/most boards/most boards you'd put a
high-end gigabit card into/. With the exception of using my own
desktop machine for driver debugging, and the cards we built in the
lab at Stanford, I've *never* used less than double-speed or double-width
(and usually both).
I suppose it depends on whether your target is a significant (3x or
more) performance increment over 100Mbit; or to deliver as close to
line-rate as possible. I've always been in the latter camp. Taking
the former case as the ``good enough'' point to stop tuning, when all
the competition is delivering (or focused on) the latter, is ... not a
good place to be.