[ofa-general] ipoib bonding problems in 1.3-beta2 and 1.2.5.4,

[Or Gerlitz]

On 12/10/07, Scott Weitzenkamp (sweitzen) <sweitzen at cisco.com> wrote:
> > Yes, that was my guess, but, my hope was that you can provide some
> > reasoning for thismethodology way of redundancy which I understand you
> > were using also for SRP HA, so can you say anything in favor of the
> > way you were working till now? As I said, this problem of failure in
> > one side enforcing a failure in the other side, and worse, when there
> > are more than two players, eg one target and N initiators, fail-over
> > in one initiator forces the target to fail-over --> forces the other
> > N-1 initiators to fail-over!?

> I think separate fabrics is a desirable, intuitive redundancy model.
> With storage each initiator fails over independently.

If the HA scheme is built on top of bonding, that is, its an IB native
file/block ULP which uses the RDMA-CM or one that runs on top of
IPoIB, then the active-backup mode of the bonding driver will not
allow each initiator to fail-over independently. For now, this is the
only mode supported for bonding/ipoib.

Or.