On Friday 14 March 2008 09:39:58 am Gerrard Geldenhuis wrote:
> > On Friday 14 March 2008 09:15:02 am Gerrard Geldenhuis wrote:
> > > anaconda adds these lines at build time to the eth devices. We have
> > > a sript in the kickstart that configures bonding and remove these
> > > lines from the eth0 and eth2 devices.
> >
> > Bonding works just fine if you leave those in there. If I recall
> > correctly, the bond takes on either a manually specified MAC address,
> > the address of whichever interface you deem primary
>
> That is correct. I add an extra parameter to specify eth0 to be primary.
> One problem should be fixed by adding the mac address back in. That is
> that randomly the devices gets swapped around so what should have been:
> eth0 and eth1 = broadcome and eth2 and eth3 = intel gets swapped around.
> The bonding driver still honours my request but it is swapped around.

Unfortunately, you do appear to have hardware that does non-deterministic
device enumeration. Might be worth trying out the kernel
parameter 'pci=bfsort' (see kernel Documentation/kernel-parameters.txt).

> The other problem we have is that the interfaces don't startup at the
> same speed. Thus even if eth0 is always pointing to the same hardware it
> might not be the current active slave in the bonding device because it
> started up slower than eth2 which then becomes the active slave.

A lot of that has to do with things outside the control of the kernel, like
negotiation with your network switch. Not really a whole lot you can do about
that on the server side as far as I know, outside of possibly bringing up the
bond with only the primary interface initially, then after a suitable delay,
bringing up the secondary.

> We want to have a predictable situation so that we know at least at
> startup that eth0 will always be the active slave.

At least in my experience, part of the idea of bonded interfaces is that you
shouldn't have to care which one is actually active...

--
Jarod Wilson
jwilson@(protected)

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