Networking in a multi-hypervisor cloud

asked 2013-11-12 09:02:37 -0600

simone.tiraboschi gravatar image

Hello, I'm trying to set a multi-hypervisor cloud. I want to have both KVM and a VMWare ESX cluster with VCenter and VMwareVCDriver.

Actually, in my plan, I want to have a single controller node and a couple of compute node: the first one runs the compute services for the KVM world and KVM itself, the second one, as far as I understood, should run the compute services for the VMWare world being connected to the VCenter node using the VMwareVCDriver.

The KVM world is perfectly working; I setup a configuration with Neutron on a different node to take advantages of Neutron capabilities in a fully segregated network configuration with per-tenant routers and private networks with overlapping capable subnets based on network namespaces. The neutron setup is based on the OVS plugin.

On the VMWare world we don't have a VMWare NSX setup so, as far as I understood, I can only setup FlatDHCPManager or VlanManager network configurations based on the old nova-network instead of Neutron that on the VMWare side should work only with Nicira NVP and Plumgrid ( , table 7.3) that are not available for me.

So I'm trying to setup an hybrid installation where the KVM compute node uses Neutron and the VCenter compute node uses nova-network. Do you think that's possible? Obviously I understand that the two kind of networking should be only lazy interoperable.

I create an Host Aggregate with an Availability Zone for the KVM world and another Host Aggregate with a second Availability Zone for the VMWare world to be associated with different flavours, is it the recommendable way?

Now the KWM world works as desiderata, unfortunately I'm not able to boot an instance on the VMWare cluster. Trying with nova boot forcing the VMWare availability zone, it reports that I should select a network id because it find multiple networks; unfortunately nova net-list reports only the neutron networks and so I'm not able to select a nova-network network to boot my VM on the VMWare cluster.

Any ideas?

thanks, Simone

edit retag flag offensive close merge delete


which was the final result of your test? and how to config the neutron.conf?Could you share here your experience on that.thanks!

yuyangwang1985 gravatar imageyuyangwang1985 ( 2017-07-07 05:03:37 -0600 )edit

3 answers

Sort by ยป oldest newest most voted

answered 2014-02-25 08:30:12 -0600

updated 2014-02-26 07:39:38 -0600

Ciao Simone,

which was the final result of your test? Did you get a full-working multi-hypervisor infrastructure with the described configuration (KVM with OVS-neutron and ESXi with nova-network).

Could you share here your experience on that.

Thanks in advance, ciao, Antonio

edit flag offensive delete link more

answered 2013-11-13 19:36:57 -0600

fifieldt gravatar image

Tricky one!

If I had to do this, I'd probably first try to use cells.

That way I could be relatively sure that each of the hypervisor worlds would be independent enough that they should take care of their own networking and not conflict. Of course, that means different network ranges, and dealing with a 2nd level of scheduler (requests first choose the cell, then the physical host) - but it will probably result in an easier time, and a better maintenance ability. You also keep your single nova-API endpoint, keystone service etc.

The problem with this approach is probably the dashboard. When you enable neutron, the way networking is done through the dashboard changes, in principle giving users access to those per-tenant routers, private subnets etc. Trying to provide that, while at the same time supporting nova-network's style will probably not work so well.

To solve that problem, you end up separating your cluster into two "regions". Which is essentially two completely separate clusters that won't have shared API endpoints, instance lists etc - not so different from two clusters.

So, I don't have a good solution for you - unless you want to drop back to nova-network for the KVM side of things...

(note also: I am not a neutron expert - there might be an neutron+vmware solution I'm not aware of)

edit flag offensive delete link more


Thanks fifieldt. I was already investigating cells and regions: a bit of segregation between the two worlds is absolutely acceptable. The two hypervisors will be used for different purposes with different service levels and so a first cell selection in the scheduling process is fully acceptable.

simone.tiraboschi gravatar imagesimone.tiraboschi ( 2013-11-14 02:40:46 -0600 )edit

As a first choice cells are preferable over region cause I'd prefer to share the API; I'll experiment about the impact of two cells with different networking architecture over horizon. I'll let you know. Ciao

simone.tiraboschi gravatar imagesimone.tiraboschi ( 2013-11-14 02:41:22 -0600 )edit

Any update on using cells? Did it work for you? What was your final solution?

kfox gravatar imagekfox ( 2014-09-05 16:29:24 -0600 )edit

answered 2017-07-07 04:56:47 -0600

yuyangwang1985 gravatar image

hi which was the final result of your test?

edit flag offensive delete link more

Get to know Ask OpenStack

Resources for moderators

Question Tools



Asked: 2013-11-12 09:02:37 -0600

Seen: 871 times

Last updated: Jul 07 '17