hi
RTFM...
I am reluctant to speak to soon but we seem to have solved the problem.

We ended up putting a laptop on each switch and then cloning traffic to
the laptop and capturing it with wireshark. We also did a tcpdump on the
server locally on each interface.

Our conclusions from this test was that a gratuitous arp was send by the
bonding driver and is seen locally on the relevant interface but is
never received/seen by the switch.

This is actually mentioned in some way in the bonding documentation.

The solution has been to set an updelay for the bond interface. That way
the switch gets a chance to get its ducks in a row before the interface
becomes active. We have found 250ms to be a good value for the updelay.
100ms have been to quick...

Btw,
The documentation for bonding makes mentioning of specifying load order
of modules.
Thus anyone know how to translate kernel 2.4 modules.conf
to kernel 2.5 modprobe.conf settings?
The command is :
add above bonding e1000 bnx2

Regards


> -----Original Message-----
> From: rhelv5-list-bounces@(protected)-
> bounces@(protected)
> Sent: 02 April 2008 10:25
> To: Red Hat Enterprise Linux 5 (Tikanga) discussion mailing-list
> Subject: RE: [rhelv5-list] bonding problems
>
> What is also notworthy to mention is that if I do a ping from the
> bonding host to the laptop then the failover is very quick.
>
> Why would eth0 -> eth2 fail over seamlessly but
> eth2 -> eth0 would not.
>
> If also tested it the other way around to ensure that it is not
> something to do with the network drivers.
>
> Regards
>

_______________________________________________
rhelv5-list mailing list
rhelv5-list@(protected)
https://www.redhat.com/mailman/listinfo/rhelv5-list