Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 19 Jun 2002 15:26:12 -0600
From:      "Kenneth D. Merry" <ken@kdm.org>
To:        "McKenna, Lee" <lmckenna@lodgenet.com>
Cc:        "'Fabien THOMAS'" <fabien.thomas@netasq.com>, paleph@pacbell.net, freebsd-net@FreeBSD.ORG
Subject:   Re: bge driver issue
Message-ID:  <20020619152612.A5514@panzer.kdm.org>
In-Reply-To: <3EA88113DE92D211807300805FA7994209149EC2@chaplin.lodgenet.com>; from lmckenna@lodgenet.com on Wed, Jun 19, 2002 at 01:46:40PM -0500
References:  <3EA88113DE92D211807300805FA7994209149EC2@chaplin.lodgenet.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, Jun 19, 2002 at 13:46:40 -0500, McKenna, Lee wrote:
> I just loaded 4.5-RELEASE, and cvsup/make world upgraded to 4.6-STABLE dated
> June 17, and have 2 machines with 3C996B-T cards using the bge driver with
> the two machines connected directly to each other using a crossover cable.
> Machines are both "home made" with Supermicro mobos with 1.2GHz CPUs.
> 
> Using /dev/zero I created a file some 1.2 GBytes in size.  I am getting 7
> MBytes/s using scp and 30 MBytes/s using ftp.  So, it looks like the max is
> about 240 Mbits/s.  This is why we affectionately call gigabit Ethernet "a
> couple-hundred megabit ethernet", and we cant wait for Ken's zero copy code
> to reach -STABLE :)

That likely won't happen until -current becomes -stable, but I won't rule
it out completely.

> This test also includes tweaking the tcp and udp send and recv space buffers
> to 512K, by the way...

That can help, but it would probably be better to use a better benchmarking
program.  scp has all the overhead of encryption to slow it down.  ftp is
notoriously bad as a benchmarking tool for the most part, because of the
way it moves data around and the socket buffer sizes it uses.

I would recommend using netperf (in /usr/ports/benchmarks).  Just crank up
the 'netserver' daemon on one end, and then run 'netperf -H hostname' on the
other end.

Try playing with the send and receive buffer sizes to see what kind of
throughput you can get.

With two machines very similar to yours, it sounds like, I can get wire
speed between two ti(4) boards (Netgear GA620T's) using jumbo frames under
a stock version of -stable.  (I can't get that with 1500 byte frames, but
that's not surprising with the Tigon 2.)

Ken
-- 
Kenneth Merry
ken@kdm.org

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-net" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20020619152612.A5514>