Bonding problem on BL460/465c

[weisstik]

Hello together,

i have three new servers and they are all running RHEL 4 AS Update 5. I have configured bonding for their NIC's. When the server start's up i always receive the error message

Jul 18 16:43:08: %C4K_EBM-4-HOSTFLAPPING: Host xx:xx:xx:xx:xx:xx in vlan XX is flapping between port Gi6/7 and port Gi7/7

on my Cisco switches. I have already implemented bonding on about 40 DL380 servers without problems. PSP 8.10 is installed on all Servers. I have already tried to downgrade the PSP to version 8.00 without any success. The problem still occurs.

Is it a Blade problem? ;o)

My /etc/modprobe.conf entries

install bond0 /sbin/modprobe -a eth0 eth1 && /sbin/modprobe bonding
options bond0 miimon=100 mode=active-backup downdelay=3000 updelay=4000


Does anyone of you have an idea for me?!

Thank you in advance.

Cheers

[Venilton Junior]

Hello,

First of all I'm not saying that you're doing it in a wrong way. Please check this HP how-to and there u may find your answer.

http://docs.hp.com/en/B9903-90050/ch05s04.html

Regards and good luck

[rick jones]

Is that flapping message on your Cisco switches in the blade chassis, or are you using pass-thru modules and connecting to external switches?

[weisstik]

@rick jones: We receive the message on the Cisco switches in the chassis.

@Venilton Junior: What should i find there? What do you think is wrong?!

[rick jones]

I'm _guessing_ that you may want to try an "lacp" mode for the bond rather than active-backup. Somewhere on the system (under /usr/ IIRC will be a "bonding.txt" file which should describe the modes in greater detail.

Or look into configuring the in-chassis-Cisco's more like the switches to which you have the DL380's connected.

[Mike Stroyan]

I wonder if the message is occuring repeatedly or if it is just a transient effect at startup.
There may be a race between the first modprobe of eth0 and eth1 and the second modprobe of bonding. There is a very small window there when eth0 or eth1 might send a packet with the same IP as the bond0 primary uses. That could cause a rapid use of the same IP from the secondary and primary NIC. You could check that eth0 and eth1 are not configured with the same IP as bond0.

[weisstik]

Hello,

I don't use any "lacp" variant.

The problem occurs when the servers starts or the network will be restartetd, everytime the bonding driver takes control over the interfaces. During this time slice the MAC is available on both ports and so the message occurs.

eth0 and eth1 don't have any IP address configured in their ifcfg-ethX file.

I have read through the documents again and again and I also have tested a lot and nothing has changed. The behaviour is still the same.

Cheers

[Court Campbell]

Is stp setup on the switch? It may be the cause of the issue.

[weisstik]

Hello,

yes, stp is configured. Both ports have spanning tree portfast enabled.

The "old" servers have the same configuration and they also have spanning tree portfast enabled and it doesn't matter.

Do you think that the problem could be the newer IOS version or switch hardware?!

Thank you in advance!

[rico_cebiche]

Your bonding type seems it is not supported by the switch. Try other type. I personally used type=6 which does not require any changes in the switch.

[Court Campbell]

You might try disabling stp on the ports for that bond. I only say that because you state that this only seem sot happen at reboots. And that has been seen as a sign of stp. I know portfast is supposed to resolve that, but it won't hurt to try.

[weisstik]

Hello,

i have updated the switch to the latest IOS release and problem disappeard ;-)

Thank you for your help!