4-port LAN strangeness

[Philip Kime]

I have an L-class with two 4-port A5506B lan cards in. I have one cable plugged into each card. Both ports are set to 100FD and same at the switch.

port 1 (Card A) doesn't come up at boot
port 2 (Card B) does

If I swap the cables, same thing so it's not the cables.

If I lanadmin -X 100HD port 1, it comes up at 100HD but it never starts properly at boot.
Same driver but STM reports a different "Software Model Number" for the cards, whatever that is ... firmware? Also, Card A reports some "Options" set (a hex number) in STM, the working Card B doesn't.

Our latest thing is to get the card replaced but any ideas much appreciated ...

[Steven E. Protter]

Without a pay for product called Auto Port Agregation, you will not be able to bring up more than one port on the card on the same network as the first port up.

This is a networking feature of HP-UX.

Example

port 1 addy 192.168.0.10 mask 255.255.255.0
That makes the network 192.168.0

Port 2 3 or 4 will not come up with network 192.168.0

What APA does is it takes all four ports and makes them seem to the network to be one, very fast port with one IP address.

SEP

[Ron Kinner]

You don't say what HPUX you are using but 11.04 has a patch PHNE_27127
and 11.11 has a patch PHKL_29468
Do you have the appropriate patch?

What does ioscan say?

Have you tried setting both sides to Autonegotiate?

Ron

[Philip Kime]

The setup is like this (HPUX 11.00 with all the latest LAN/card patches):

port 1 (Card A) is set to be up at boot with a certain IP

port 2 (Card B) is set to be down at boot but with the same IP (don't ask me - it's their idea of "redunancy" without paying for APA - manual failover ...)

This is how it's configured on every production box and it works.

At boot:

port 1 (Card A) set to come up, port 2 (Card B) down, nothing - the IP isn't reachable until I do a lanadmin -X 100HD on port 1

BUT

port 2 (Card B) set to come up at boot, port 1 (Card A) down - everything is fine

If I switch port 1 (Card A) to port 2 (Card A), exactly the same problem.

so this seems to me like there is something wrong with Card A ...

[Steven E. Protter]

Phillip,

As far as I know ServiceGuard standby lan card or APA is the only way to accomplish what you are trying to accomplish.

I do have some custom hp-ux shell scripts running on my personal hp9000 server that check other IP addresses on the network and if they find a node is down, attempt to take over the IP address and send an email that makes my cell phone beep. But thats clearly a custom job and not any kind of scheme that I have heard HP supports. I will say my implementation is pretty nifty(Geek Alert!).

Both APA and ServiceGuard are licensed pay for products.

If your goal is failover a scripted failover is possible.

SEP

[Philip Kime]

I realise that it should be done with some proper software or scripts - but this is exactly what they do here - there is a script to detect lan1 failure (port 1, Card A) which then takes this port down and brings lan5 (port 1, Card B) up. It's not hard to do, as you say but surely when in /etc/rc.config.d/hpbtlanconf, lan1 is marked "up" and lan5 "down" when the IPs are the same, it would be fine? The weird thing is, when lan5 is marked "up" and lan1 "down", it boots perfectly fine and the network is visible! To summarise, on boot:

lan1 (port 1, Card A) = up
lan5 (port 1, Card B) = down
RESULT = NO NETWORK until flip lan1 to 100HD

lan1 (port 1, Card A) = down
lan5 (port 1, Card B) = up
NO PROBLEMS

lan2 (port 2, Card A) = up
lan5 (port 1, Card B) = down
RESULT = NO NETWORK until flip lan2 to 100HD

Same cards, same IPs, same drivers, same switch settings etc. I don't think that this is a forbidden configuration without APA or whatever since only one port with the IP is ever up and it works if that port is lan5, just not when it's any lan port in Card A. I did renumber the lan instances with ioinit earlier because there was a problem with a gelan interface hanging around as lan1 in /etc/ioconfig that had been physicall removed. But this shouldn't have done anything nasty should it?

[rick jones]

when you changed instance numbers did you also go-in and edit any of the btlan config files to match the changed instance numbers?

wrt to ports even coming-up with IPs in the same subnet, there is nothing in the transport that _prevents_ someone from assigning IPs in the smae subnet to multiple plysical (from the transports perspective) NICs. what happens though is that while all interfaces configured into the same subnet will be used for inbound, only one will be used for outbound, and this is not typically what folks want.

[Philip Kime]

Yes - hpbtlanconf is fine - this was all changed corresponding to the new instance numbers.

The thing is, we never want or have both cards up at the same time so there is never any case where we have two cards up on the same subnet.

[Philip Kime]

Solved - netconf is a bit stupid - the "down" lan5 must be completely commented out in netconf - you can't have an entry for it and say "status=down" or you get these problems ...