Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 21 Mar 2001 20:47:26 +0100 (MET)
From:      Helge Oldach <Helge.Oldach@de.origin-it.com>
To:        conrad@th.physik.uni-bonn.de (Jan Conrad)
Cc:        gordont@bluemtn.net, bright@wintelcom.net, dillon@earth.backplane.com, rdm@cfcl.com, freebsd-stable@FreeBSD.ORG
Subject:   Re: NFS performance
Message-ID:  <200103211947.UAA03351@galaxy.de.cp.philips.com>
In-Reply-To: <Pine.BSF.4.33.0103211856270.1867-100000@merlin.th.physik.uni-bonn.de> from Jan Conrad at "Mar 21, 2001  6:57:39 pm"

next in thread | previous in thread | raw e-mail | index | archive | help
Jan Conrad:
>On Wed, 21 Mar 2001, Gordon Tetlow wrote:
>> On Wed, 21 Mar 2001, Jan Conrad wrote:
>>
>> > ifconfig fxp0:
>> > fxp0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500
>> >         inet 131.220.161.127 netmask 0xfffffc00 broadcast 131.220.163.255
>> >         ether 00:90:27:1c:f3:79
>> >         media: 100baseTX status: active
>> >         supported media: autoselect 100baseTX <full-duplex> 100baseTX
>> > 10baseT/UTP <full-duplex> 10baseT/UTP
>> >
>> > server:
>> > fxp0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500
>> >         inet 131.220.161.121 netmask 0xfffffc00 broadcast 131.220.163.255
>> >         ether 00:02:b3:1f:f8:c5
>> >         media: 100baseTX status: active
>> >         supported media: autoselect 100baseTX <full-duplex> 100baseTX
>> > 10baseT/UTP <full-duplex> 10baseT/UTP
>>
>> As an aside you probably want to force to 100baseTx <full-duplex> if you
>> are using a switch.
>
>I'll try to convince the administrator of that switch (very difficult)

There is a chance that you could do it without bothering the admin.

The first thing to do is to look at the collisions (netstat -in). If
they are huge (say, >5% of packet total), the switch probably is already
at full-duplex, but your machine didn't recognize. In that case you
could just ifconfig your NIC and see how it goes. If the collision
counter doesn't change even under load, you're done.

Otherwise you could try to *force* the switch to "100BaseTX
<full-duplex>" using the mediaopt option on ifconfig. Make sure to have
the link physically disconnected for a second from the switch, so that
the switch observes a trigger to re-negotiate. If you are lucky, the
switch will understand your NIC's full-duplex offering and align itself
properly.

In any case you definitely need to verify that the switch went
full-duplex, in particular if you can't look at the port setting by
means of LEDs or a management station. A simple test would be a flood
ping (-f option) to another machine at the same switch (or, better, to
the switch's internal management address). If you see significant packet
loss, the switch probably didn't grok it. In that case there is no
option but to revert to half-duplex.

Background: 10/100 auto-negotiation usually works reliably, but duplex
negotiation doesn't (and conceptually can't). This is a Bad Thing: While
failing 10/100 auto-negotiation will be immediately visible because
the NIC just doesn't work at all, failing duplex negotiation won't be
immediately obvious. In fact you probably will see all sorts of strange
performance issues if one party is at full and the other is at half, but
they won't be obviously pointing to the network cable. So be warned.

As a side note I've seen good auto-negotiation results with fxp NICs and
assorted Catalysts.

Helge

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




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200103211947.UAA03351>