NIC Speed setting change upon reboot

[Peter Chokola]

We have an Alpha Server running Tru64 and TruCluster that was given to us to support, and we are having a hard time setting the nic card duplex/speed/autonegotiation settings so that they are not lost after a reboot.
On Linux, HP, etc. there is a file that needs to be edited so that the setting is retained. What is the file on Tru64?

[Ivan Ferreira]

Add to /etc/inet.local the following:

/usr/sbin/lan_config -i -a 0 -s 100 -x 1


In this example, the interface is configured to 100 full duplex autonegotiation disabled.

[Steven Schweda]

The system model and the type of NIC might be
interesting. In many cases, the SRM console
is used to set durable NIC characteristics.

>>>show e*_mode


Note: "nic card" = network interface card card.

See also:

http://antinode.org/complaints/aer.html

[Peter Chokola]

Here is some more background.

I checked the /etc/inet.local file and indeed the settings that you mentioned are there.

/usr/sbin/lanconfig -a 0 -i ee0 -m utp -s 100 -x 1

/usr/sbin/lanconfig -a 0 -i ee1 -m utp -s 100 -x 1

I am not familiar with Tru64 or TruClustering environments, so I do not understand how there are settings for two interface cards, when there are actually four cards. There is an ee0 and a ee1 nic card in System A and an ee0 and a ee1 nic card in System B. It is my understanding that these machines share the same files. Does the ee0 settings above affect the ee0 cards in each server, and the same for ee1?

I ran the "hwmgr get attribute" command on each cluster member, and all four nic cards show (Four separate mac addresses), and each one is set up as Full Duplex= 1, Autoneg_enable= 0, and Media Speed= 100MBS.

Now, here is the problem we are having. When I boot the two members of the cluster, one server comes up half duplex and the other full duplex. This occurs specifically on the lan cards that are used for the communication between the servers, which causes major problems as you can imagine. I just discovered today that both lan cards have now magically switched to full duplex after some unknown period of time (After a day or so?). So that is what we do not understand. We put a hub between the two servers to verify that this was occurring.

As far as the question of the equipment:
Server:Alpha Server ES40 (MODEL DA-64CAA-EA).
Four Lan Cards: Intel 82559.
Someone at Hp told me that this is sometimes referred to as an EMS12000 type system or similar.

[Rob Leadbeater]

Hi Peter,

Although some files in TruCluster are shared between all nodes, some are stored in cluster specific locations.

I've not got a system in front of me to check at the moment, but you'll probably find that /etc/inet.local is actually a symbolic link to ../cluster/members/memberX/etc/inet.local

where X is 1 or 2 for each of the two systems A or B.

You should check that the settings are set correctly in both of these files.

You should also make sure that the settings are set correctly at the hardware level on both nodes when at the P00>>> prompt.

The previously mentioned show e*_mode command will show you what they're set to.

Hope this helps,

Regards,

Rob

P.S. Steven: I thought NIC was Network Interface Controller so NIC Card isn't necessarily incorrect !!

http://en.wikipedia.org/wiki/Network_card

[Peter Chokola]

Thank you for explaining the symbolic link method for the cluster. I checked, and the inet.local files are the same on both nodes.

Perhaps the problem indeed has to do with the settings at the PO prompt. I gather that I need to be at the PO prompt to run the e*_mode command because I can not get it to work, or I do not know the command path? I can not boot the server to get to the PO prompt because it is heavily being used. Maybe I will remember to check it the next time something goes wrong.

[Al Licause]

RE: console settings.

For any network card other than the older DEC tulip chips, seen at the operating system level as tu devices, the console settings never come into play.

None of the drivers for any of the non-DEC type devices in including ee, bcm, and alt actually read the console settings.

While you may find postings to the contrary in this and other forums, please know that engineering from DEC, Compaq and HP have confirmed repeatedly that only tulip drivers read the console settings. Any resulting settings for any other drivers and console configurations are purely coincidence.

From your description, it would seem that while you specifically disabled autonegotiation in inet.local, if the hub or switch ports were not set to match that of the host, a mismatch will occur and the communications will either be slow or not work at all.

In all cases, the settings for the network cards on the host, must match that of the corresponding switch settings.

[Peter Chokola]

Actually, we put the switch in the system after we noticed a duplex mismatch between the two servers with the direct ethernet cable connection. We put the switch in just to verify with the led lights that there was a mismatch, and we could see if any of the changes we made would show up as fixing the problem where one system was half duplex.

[Ivan Ferreira]

You can verify the current network speed and duplex with:

hwmgr get attr -cat network

You can run the lan_config command anytime you want to modify the settings.

Remember to check this page also:

http://forums1.itrc.hp.com/service/forums/helptips.do?#28

[Moderator edit: The above link is no longer valid.]

[Peter Chokola]

Yes I already verified the settings with the hwmgr. I thought that maybe this was going to be something obvious, but it is starting to look like this will be an obscure fix. Someone from Compaq is probably going to have to come in and look at the system. I will leave the thread open a little longer if someone has any ideas.

[Al Licause]

If you can't get a resolution to your problem through this forum, please consider logging a call to the HP Customer support center if you have a support contract.

Even if you don't, many times they'll let you in anyway....it's worth a shot.

And if you do get to talk to a specialist, they can ask for specific files that may help resolve the issue.

[Peter Chokola]

Unresolved. Still waiting for a support contract to be reactivated.

[Bill Finley]

Peter,

I am have the problem to have with my ES40s. The switch the NICs are attached to are set at 100Mbs/Full Duplex. The NICs do not set it to Full Duplex until Unix comes along and run lan_config via the inet.local file. This is late in the boot process and impacts boot performance. I will try to pursue an fix. We still have a valid support agreement. I wonder why the driver manufacturer would ignore the console settings.

Bill Finley

[Steven Schweda]

> I am have the problem to have with my ES40s.

Sounds like a good opportunity for a fresh
thread.

> The NICs do not set it to Full Duplex [...]

NIC model(s)?

Do they have modes set by the console?