slow lan

[Todd McDaniel_1]

I have an issue where my network admin made some changes to our network and now since last week my admin has been complaining about backups taking too long and having to kill them before the online day begins. They say that somehow the network is nego to half instead of remaining at 100FD which is configured.

The network guy wants to alter my lan configuration to go from 100FD to 1GB because he says that 1GB setting doesn't offer half duplex so the connection should stay at FD instead of half, even though both the switches and the servers are configured for 100FD.

He says that somehow they are still nego to half, which I am hard pressed to understand, since all hardware has been hardcoded at 100FD.

Here is his direct quote:

+++++++++++++++++++++++++++++++++++++++++++
The network is set to 100/full, the nic is set to 100/full but the link is established at 100/half. Half duplex is the issue. We don’t have a speed problem we have a duplex problem. Since gigabit cannot operate in halk duplex it is worth a try to set the elextra nic to 1000/full and see if we can establish a common link between the network switch and the server nic.
++++++++++++++++++++++++++++++++++++++++++
unquote


here is my lan card from ioscan:

Class I H/W Path Driver S/W State H/W Type Description
===================================================================
lan 0 0/1/2/0 igelan CLAIMED INTERFACE HP PCI 1000Base-T Core


My question is what would I need to reconfigure, if anything to even attempt this change. what changes to the netconf file?

MY next question, will any of this matter if the issue isn't on the server? I dont believe it is.

[Sandman!]

Agree that it does not seem to be related to the server esp. since changes were made to the network. Post the kind of server and NIC card that are in question and the output of...

# ioscan -funC lan

[Tim Nelson]

What does your lanadmin -x 0 say ?

[Jeff_Traigle · Accepted solution]

Not sure why you'd want to force things to 100FD if the NIC and switch both support gigabit. You can't force 1000FD. To run at gigabit, you must configure the card to autonegotiate. (The driver doesn't give you an option even.)

The netconf file doesn't control this. It's the /etc/rc.config.d/hpigelanconf file where you set speed, duplex, and autonegotiate options for the igelan cards.

[Todd McDaniel_1]

#ioscan -funC lan
Class I H/W Path Driver S/W State H/W Type Description
===================================================================
lan 0 0/1/2/0 igelan CLAIMED INTERFACE HP PCI 1000Base-T Core

#lanadmin -x 0
Speed = 100 Full-Duplex.
Autonegotiation = Off.

#uname -a
HP-UX elektra B.11.11 U 9000/800 1193809610 unlimited-user license

#model
9000/800/rp3440

[A. Clay Stephenson]

If you want to establish a 1GB/s connection then BOTH ends MUST be set to auto-negotiate. You can hard-set both ends to 100FD or 100HD if you like BUT if one end is hard-set to 100FD the other end MUST also be set to 100FD. A sure-fire recipe for just your sort of problem is to hard-set one end and leave the other end set to auto-negotiate.

You can't really say that the issue isn't on the server because whatever the setting is on the server (auto-negotiate, 100FD, 100HD, 10HD) it MUST match that of the corresponding ethernet switch port. The least-evil answer in most cases with modern boxes is to set the card (and corresponding swith port) to auto-negotiate.

[Tim Nelson]

Related to Jeff's comment.

It was always a good thing to force the older 10/100 cards to a specific speed and duplex.

I believe it is recommended not to force 1000GB cards but leave them in auto-neg mode. The switch needs to be set to auto also.

[Todd McDaniel_1]

Jeff,

Everything is set to 100fd and was that way before I got these boxes. Only the network side changed last Tuesday.

I believe the card is 1g based on the ioscan, but I am not sure if that is correct.

All I do know is, I don't think changing the host will fix the issue.

Somehow the network is nego to half Duplex, and my network guy seems to think changing the host will help it somehow.

[Todd McDaniel_1]

ignore my last note... :)

I just read some of your replies about hardcoding 100FD.

I see the light, but our systems were hard coded before I got them and seemed to work normally.

I have only had these boxes a few weeks, so havent had the chance to really go over them.

[Jeff_Traigle]

What did the network guys change? It's seems from the quote that they think they still have the port set to 100FD, but it sure sounds like they changed it to autonegotiate, which will cause issues like you're having since the NIC is set to 100FD. Never trust those network guys. ;)

The NIC is definitely gigabit. So, if they actually set their switch port to autonegotiate, as I suspect they did, and you change the NIC to also autonegotiate, then it should indeed fix the problem and you'll be running at 1000FD.

[Todd McDaniel_1]

Okay thanks guys. Whew!!

I need to take a crash course in networking :)


BTW, can I change that setting online without rebooting?

[A. Clay Stephenson]

Yes, you can run the lanadmin online and this will trigger a reset of the card and a loss of connection for a few seconds. It's probably wise to execute this command from the console so that you don't rely on a network connection. I like to be able to get myself out of trouble almost as fast as I got myself in. You will also need to change the configuration file in /etc/rc.config.d so that the changes will be preserved across a reboot. However, don't do anything until you KNOW how the corresponding switch port is set and if it were me it would be auto on both ends.

[Todd McDaniel_1]

so just to be clear.

I run a reset from the lanadmin tui? and that will re-read the /etc/rc.config.d/hpigelanconf or how will the card know to autoneg?

[A. Clay Stephenson]

No, you are going to execute lanadmin probably as "lanadmin -s auto_on NM_ID".

[A. Clay Stephenson]

Ooops that should be: No, you are going to execute lanadmin probably as "lanadmin -S auto_on NM_ID". Man lanadmin for details.

[Todd McDaniel_1]

NM_ID I take it is the PPA number?

#lanadmin -S auto_on 0
Old Speed = 100000000
Unable to change speed: errno = 22

[Todd McDaniel_1]

I guess with that error I will have to reboot to change it.

[Sandman!]

Wait...don't reboot. The lanadmin(1M) command should be:

# lanadmin -X auto_on 0

[Bill Hassell]

Just a comment on why the 100 Mbit link is going to HD. IF the two ends (NIC and switch) are set to auto-negotiate, then (assuming non-ancient hardware) the link will be 100 FD. BUT: if the cable length falls into the 27-35 meter range, then the link pulses can be confused by certain LAN hardware chips and the rule is: mismatch = half duplex. So if the switch and NIC worked fine a couple of weeks ago, the switch was probably changed to autonegotiate or perhaps the cable was changed.

100 Mbit is the only link speed where FD, HD and autonegotiate can be selected. At 1 Gbit, the link *MUST* be set to autonegotiate on both sides. The cable range mentioned above has no effect on 1 Gbit links.

[V. Nyga]

Hi,

to check if the mismatch is on your box, you can check the lan statistics:
'lanadmin' - 'lan' - 'dis' will show you your settings.
After this 'Press to continue' appears, after you'll see the network stats - if there are 'FCS Errors' (not equal zero) then you've the problems at the network connection of your box.

You can clear the stats, to see if changes had success.

HTH
Volkmar
and yes - *don't believe the network guys* ;-)

[rick jones]

Your networking guys are living in mystical world of folklore and history not supported by present conditions.

Attached is some boilerplate about how autoneg is supposed to work.

[Matti_Kurkela]

Some Cisco gigabit switch modules (for Catalyst 6500) behave in an interesting manner.

The ports in those modules *can* be forced to 1000/full, but the actual mechanism of this forcing is very different from how it used to be with the 100baseTX modules.

When the autonegotiation is disabled in a traditional 100baseTX network port, the port will not send nor respond to autonegotiation messages.

When a 1000baseT network port is forced to 1000/full, it will still respond to autonegotiation messages... but the content of those messages is essentially "1000/full only. Take it or leave it."

Some network admins seem to be unaware of this difference.

MK