[ovirt-users] Best practices for LACP bonds on oVirt

Vinícius Ferrão ferrao at if.ufrj.br
Tue Jul 4 08:27:14 UTC 2017


> On 4 Jul 2017, at 02:49, Yedidyah Bar David <didi at redhat.com> wrote:
> 
> On Tue, Jul 4, 2017 at 3:51 AM, Vinícius Ferrão <ferrao at if.ufrj.br> wrote:
>> Hello,
>> 
>> I’m deploying oVirt for the first time and a question has emerged: what is the good practice to enable LACP on oVirt Node? Should I create 802.3ad bond during the oVirt Node installation in Anaconda, or it should be done in a posterior moment inside the Hosted Engine manager?
> 
> Adding Simone for this, but I think that hosted-engine --deploy does
> not know to create bonds, so you better do this beforehand. It does
> know to recognize bonds and their slaves, and so will not let you
> configure the ovirtmgmt bridge on one of the slave nics of a bond.
> 
>> 
>> In my deployment we have 4x GbE interfaces. eth0 and eth1 will be a LACP bond for management and servers VLAN’s, while eth1 and eth2 are Multipath iSCSI disks (MPIO).
> 
> You probably meant eth2 and eth3 for the latter bond.
> 
> This is probably more a matter of personal preference than a result of
> a scientific examination, but I personally prefer, assuming that eth0
> and eth1 are managed by a single PCI component and eth2 and eth3 by
> another one, and especially if they are different, and using different
> kernel modules, to have one bond on eth0 and eth2, and another on eth1
> and eth3. This way, presumably, if some (hardware or software) bug hits
> one of the PCI devices, both bonds hopefully keep working.

It’s one single card with 4 interfaces. It came onboard on the IBM System x3550 M4 servers that I’m using and they are Intel based, don’t remember exactly which chipset. Anyway this is interesting, I really avoid mixing different controllers on bonding to keep things stable, but you’ve a point.

> Just my two cents,
> 
>> 
>> Thanks,
>> V.
>> 
>> _______________________________________________
>> Users mailing list
>> Users at ovirt.org
>> http://lists.ovirt.org/mailman/listinfo/users
> 
> 
> 
> -- 
> Didi



More information about the Users mailing list