← Back to team overview

openstack team mailing list archive

Re: multi_host networking, but not on all nodes?


My two cents...

The current multi-host mode is really making the assumption that if the NC
running in the dom0 is gone, all the VMs are likely screwed anyway.

When you are talking about a middle-ground, you'll need to handle NC
failures and load-balance among the NCs. You'll also need to worry about
the traffic, because the placement of NCs on a subset of nodes becomes

What you really need is a scalable NATter array, which kinda load-balance
and fail-over across projects. I'm trying to prototype this now with my
colleague actually :-)


On Tue, Feb 7, 2012 at 4:27 PM, Nathanael Burton <
nathanael.i.burton@xxxxxxxxx> wrote:

> With the default networking there's a single nova-network service.
> With the --multi_host option, 'set_network_host' sets every instance
> to use their host as the nova-network node, effectively requiring
> nova-network to run on every nova-compute.  The multi_host mode
> greatly helps HA and consolidates fault domains, but at the cost of
> increased complexity and IP sprawl when using the VLAN networking
> model, as each host in the zone now has to have an IP on every VLAN.
> What I think I'm looking for is a middle ground where you can run
> multiple nova-network nodes, but not equal to the number of compute
> nodes.  Basically a similar ability as implemented with the
> nova-volume service; the ability to scale the nova-network nodes
> independently from the computes.  The big downside is that you no
> longer have the benefit of combined fault domains (network/compute).
> Is any of this possible today?  Does Quantum with OpenvSwitch handle
> any of this either?
> Thoughts?
> Thanks,
> Nate
> _______________________________________________
> Mailing list: https://launchpad.net/~openstack
> Post to     : openstack@xxxxxxxxxxxxxxxxxxx
> Unsubscribe : https://launchpad.net/~openstack
> More help   : https://help.launchpad.net/ListHelp