SEO

I've encountered something strange on one of our T1's, after activating
the interface, adding ip address and routing, approx. one out of ten
pings from other servers works. There's also this strange message in
/var/adm/messages on the T1 once the NIC is activated:

Jun 2 15:19:10 lab-t1 hme: [ID 786680 kern.notice] SUNW,hme1 : External
Transceiver Selected.
Jun 2 15:19:10 lab-t1 hme: [ID 786680 kern.notice] SUNW,hme1 : 0 Mbps
Full-Duplex Link Up

The switch port is set to 100/full, I've ran all the neccessary ndd -set
commands to set the T1 to 100/full and disable all other speed/duplex
settings, tried different cables and different ports on the switch.

Monitoring the switch shows that the pkt count increases if I ping the
server, but there are no collisions or drops.

I have made a simple rc2.d/S01 duplex script which sets the hme's to
100-full in the following way:
#! /bin/sh
#

echo "Forcing interface hme0 to 100 Mbps full duplex"
ndd -set /dev/hme instance 0
ndd -set /dev/hme adv_100fdx_cap 1
ndd -set /dev/hme adv_100hdx_cap 0
ndd -set /dev/hme adv_10fdx_cap 0
ndd -set /dev/hme adv_10hdx_cap 0
ndd -set /dev/hme adv_autoneg_cap 0

echo "Forcing interface hme1 to 100 Mbps full duplex"
ndd -set /dev/hme instance 1
ndd -set /dev/hme adv_100fdx_cap 1
ndd -set /dev/hme adv_100hdx_cap 0
ndd -set /dev/hme adv_10fdx_cap 0
ndd -set /dev/hme adv_10hdx_cap 0
ndd -set /dev/hme adv_autoneg_cap 0

So is 0 Mbps Full-Duplex Link Up is an indication of a flakey NIC, or is
there something else?

--
Stig Bull
| remove .no.spam from my email address to reply by mail |
No animals were hurt or killed in the process of creating this
electronic message. To reduce download time, this message is made of
100% recycled bytes.

Stig Bull wrote:

> The switch port is set to 100/full, I've ran all the neccessary ndd -set
> commands to set the T1 to 100/full and disable all other speed/duplex
> settings, tried different cables and different ports on the switch.

you should let the NIC and the switch autonegociate instead of forcing them.

Stig Bull <stig.bull.no.spam@broadpark.no> writes:
> echo "Forcing interface hme0 to 100 Mbps full duplex"
> ndd -set /dev/hme instance 0
> ndd -set /dev/hme adv_100fdx_cap 1
> ndd -set /dev/hme adv_100hdx_cap 0
> ndd -set /dev/hme adv_10fdx_cap 0
> ndd -set /dev/hme adv_10hdx_cap 0
> ndd -set /dev/hme adv_autoneg_cap 0

You've disabled autonegotiation (adv_autoneg_cap 0), so setting the
autonegotiation parameters (the other adv_*_cap values) won't actually
do anything useful.

I'd strongly recommend turning autonegotiation back on. It's a
required part of the 802 standards, and you can get all sorts of
flakey behavior if you turn it off.

--
James Carlson, KISS Network <james.d.carlson@sun.com>
Sun Microsystems / 1 Network Drive 71.234W Vox +1 781 442 2084
MS UBUR02-212 / Burlington MA 01803-2757 42.497N Fax +1 781 442 1677

On Fri, 03 Jun 2005 10:39:52 -0400, James Carlson wrote:

> Stig Bull <stig.bull.no.spam@broadpark.no> writes:
>> echo "Forcing interface hme0 to 100 Mbps full duplex"
>> ndd -set /dev/hme instance 0
>> ndd -set /dev/hme adv_100fdx_cap 1
>> ndd -set /dev/hme adv_100hdx_cap 0
>> ndd -set /dev/hme adv_10fdx_cap 0
>> ndd -set /dev/hme adv_10hdx_cap 0
>> ndd -set /dev/hme adv_autoneg_cap 0
>
> You've disabled autonegotiation (adv_autoneg_cap 0), so setting the
> autonegotiation parameters (the other adv_*_cap values) won't actually
> do anything useful.
>
> I'd strongly recommend turning autonegotiation back on. It's a
> required part of the 802 standards, and you can get all sorts of
> flakey behavior if you turn it off.

Yes, one in particular is Cisco switches will not see an autonegotiate
preamble on the wire and will default to 100-half if they are set to
autonegotiate. I
heard alot of whining from idiots on that one (la, la, la, Solaris sucks,
your machine is broken). Tell them what I found and what the fix was,
then they say, "Oh yeah, that guy moved offices a week ago."

- Scott

Scott Packard <scottp@15.usenet.us.com> writes:

[hme interface]

Yes, one in particular is Cisco switches will not see an autonegotiate
preamble on the wire and will default to 100-half if they are set to
autonegotiate.

I don't know the full story, but disabling auto-negotiation for
a Sunswift add-on interface was our solution for negotiation problems.

This is an ancient hme device (100 Mb/s, and SCSI) added to an
Ultra-1. Other Suns on the same wire have no problems with
autonegotiation. For the sunswift card, setting both ends 100Mb/s
full worked. The troublesome behavior was repeatable with multiple
switches.

--
--Darrel Hankerson dhankers@cacr.math.uwaterloo.ca

Thank you all for your reply, we have switched autoneg off for all of
our servers since they caused all kind of problems with our old Cisco
switches. However we have switched to HP switches now, so you still
think I should turn on autoneg?

The problem with hme1 was that local-mac-address? was set to false in
OPB - I think this is by default, so both hme0 and hme1 shared the same
MAC address. Setting it to true and rebooting the server fixed the
problem.

--
Stig Bull
| remove .no.spam from my email address to reply by mail |
No animals were hurt or killed in the process of creating this
electronic message. To reduce download time, this message is made of
100% recycled bytes.

In article <MPG.1d0f5a4338f285399896aa@news.banetele.no>,
stig.bull.no.spam@broadpark.no says...

Just to elaborate on my previous post, autoneg is turned off on the
switch, all ports are set to 100-full. Should I still enable autoneg on
the server NIC's?

--
Stig Bull
| remove .no.spam from my email address to reply by mail |
No animals were hurt or killed in the process of creating this
electronic message. To reduce download time, this message is made of
100% recycled bytes.

On Tue, 07 Jun 2005 11:16:03 +0200, Stig Bull wrote:

> In article <MPG.1d0f5a4338f285399896aa@news.banetele.no>,
> stig.bull.no.spam@broadpark.no says...
>
> Just to elaborate on my previous post, autoneg is turned off on the
> switch, all ports are set to 100-full. Should I still enable autoneg on
> the server NIC's?

My thinking is: Will you remember the settings for all eternity?
It's working fine, but Cisco upgrades here have caused the loss
of configuration data. I'm sure the same thing could happen to your
servers. When that happens then the ports aren't going to match each
other.
It's simpler to leave things in their default settings, and then if
they won't play together document the non-default settings everywhere
you can think of.

Would it help to upgrade your server NICs?

- Scott

In article <pan.2005.06.07.19.34.45.499418@%HASH%.usenet.us.c om>,
scottp@15.usenet.us.com says...

> > Just to elaborate on my previous post, autoneg is turned off on the
> > switch, all ports are set to 100-full. Should I still enable autoneg on
> > the server NIC's?
>
> My thinking is: Will you remember the settings for all eternity?

Forced 100-Full has been an IT policy since before I started working
here, the company had all kinds of problem with Cisco switches and
autoneg.


--
Stig Bull
| remove .no.spam from my email address to reply by mail |
No animals were hurt or killed in the process of creating this
electronic message. To reduce download time, this message is made of
100% recycled bytes.

Stig Bull <stig.bull.no.spam@broadpark.no> wrote:
> Forced 100-Full has been an IT policy since before I started working
> here, the company had all kinds of problem with Cisco switches and
> autoneg.

Exactly. It's a policy which made sense many, many years ago. Today
it causes people more pain than any other single network problem.

Are you still running Token Ring and CDDI? They too were many companies
IT policy years ago...

Scott