We just changed this up a little this week. We split our traffic into 2 bonds, 10GB mode 1
as follows:
Guest vlans, managment vlan (including some NFS storage) -> bond0
Migration layer 2 only vlan -> bond1
This allowed us to tweak the vdsm.conf to speed up migrations without impacting management
and guest traffic. As a result we’re currently pushing about 5Gb on bond1 when we do live
migrations between hosts.
-Patrick
On Jul 28, 2015, at 1:34 AM, Alan Murrell <lists(a)murrell.ca>
wrote:
Hi Patrick,
On 27/07/2015 7:25 AM, Patrick Russell wrote:
> We currently have all our nics in the same bond. So we have guest
> traffic, management, and storage running over the same physical
> nics, but different vlans.
Which bond mode do you use, out of curiousity? Not sure I would go to this extreme,
though; I would still want the physical isolation of Management vs. network/VM traffic vs.
storage, but just curious which bonding mode?
Modes 1 and 5 would seem to be the best ones, as far as maximising throughput. I read an
article just the other day where a guy detailed how he bonded four 1Gbit NICs in mode 1
(with each on a different VLAN) and was able to achieve 320MB/s throughput to NFS
storage.
As far as the storage question, I like to put other storage on the network (smaller NAS
devices, maybe SANs for other storage) and would want the VMs to be bale to get at those.
Being to use a NIC to carry VM traffic for storage as well as for host access to storage
would cut down on the number of NICs I would need to have in each node.
-Alan
-Alan