[ovs-discuss] Bonding problems with redundant physical switching

Ben Pfaff blp at nicira.com
Wed Sep 15 16:47:38 UTC 2010


On Wed, Sep 15, 2010 at 04:39:50PM +0000, Jesse Gross wrote:
> On Wed, Sep 15, 2010 at 4:29 PM, Ben Pfaff <blp at nicira.com> wrote:
> > On Wed, Sep 15, 2010 at 12:17 AM, Christian Fischer
> > <christian.fischer at easterngraphics.com> wrote:
> >> You sad that I can use OVS with Linux bonding, how is that to do?
> >> Simply loading the bonding kernel module causes segmentation fault, in my
> >> case. Can/Must I compile OVS with linux bonding support?
> >
> > Hmm, it looks like the bonding module depends on the bridge module, at least
> > on some Linux kernel versions.  Indeed, it may not be possible to use the
> > Linux bonding at the same time as Open vSwitch for Linux before 2.6.36.
> 
> Actually Linux bonding should not be used with Open vSwitch even if it
> were not for this interaction with the bridge module.  Bonding has a
> number of interesting interactions with the learning table and Open
> vSwitch has code to handle this but only if the bonding is handled by
> OVS itself.  As a result, Linux bonding may work (assuming a new
> enough kernel version), however, you are likely to run into a number
> of difficult to diagnose corner cases.

I was under the impression that those corner cases were problems
specific to the type of bonding that OVS already implements; that is,
bonding where the first-hop switch was unaware of and not participating
in the bond.  Doesn't LACP, etc. avoid that kind of problem, acting much
more like a single link than SLB does?




More information about the discuss mailing list