Re: NIC bonding on pSeries with dual VIOS
On Thu, Feb 12, 2009 at 11:40:03PM +0100, Frank Fegert wrote:
> Hello all,
>
> i've got Debian (testing) running on a DLPAR on a pSeries 9117-570.
> The system gets several NICs from two VIO servers, which in turn are
> connected to different switches for redundancy. Two NICs within each
> subnet are paired via bonding devices.
> Now the problem:
> - If i choose 'balance-rr' as bonding mode our network guys complain,
> because the same MAC appears on two different switchports that be-
> long to the same VLAN. (Because the bonding interfaces MAC gets
> forced onto each slave eth-interface.)
If you uipgrade to a newer (sorry) kernel then you'll get a failover_mac option
to the bonding driver. This will allow you to specify that the MAC address for
outbount packets should be the hardware MAC address. Thsi will stop your
network admains complaining ;P
However it will not work out of the box. You'll also need to ensure that both
underlying network devices are in promisicuous mode. This shoudl give you the
failover you want
Of course it may not be suitable in your environment.
> - 'balance-alb' would be an alternative, since it uses the MAC of
> the slave eth-interfaces instead of the one from the bonding inter-
> face. Unfortunately 'balance-alb' does not seem to be supported
> with the ibmveth driver.
Hmm can you provide some more details here? I can't see a reason why this
would be so?
> - 'active-backup' would also be an alternative, but then i can't
> reach other LPARs on the same machine which currently have their
> active interface on another VIO. LPARs with their active interface
> on the same VIO can be reached. This is strange, because the other
> LPARs (all AIX) use NIB (also active-backup NIC bonding) and can
> all reach eachother no matter on which VIO their active interface
> is.
>
> Is the 'active-backup' problem a limitation of the ibmveth driver, or
> am i missing something here?
Not sure about your active-backup issuses, I've never used that mode.
Yours Tony
Reply to: