Ask Your Question

VMWare hypervisor as compute node?

asked 2014-06-13 20:36:10 -0600

Shardy gravatar image

I've set up a lab environment using KVM on the compute node. But in production I'll probably need to switch to using VMWare. I'm not exactly clear on whether I still need to install compute node(s) to control VMWare or if the controller node can talk directly to the VMWare hypervisors?

In other words can Openstack treat VMWare hypervisor as a compute node directly or does it need an intermediary compute node to talk to it?

edit retag flag offensive close merge delete


dbaxps gravatar imagedbaxps ( 2014-06-13 21:27:58 -0600 )edit

4 answers

Sort by ยป oldest newest most voted

answered 2014-06-14 02:09:26 -0600

SGPJ gravatar image

For network related issues: please refer to

edit flag offensive delete link more

answered 2014-06-16 05:09:45 -0600

Venu Murthy gravatar image

Hello Shardy,

Just like right now you are using Compute to fire API requests to KVM using the Libvirt driver (setup in the nova.conf) similarly you will continue to use the Compute to fire API requests to the VMware hypervisor using VMWare's drivers, which are already provided in the nova installation.

And in a prod environment you will see that vCenter is used for orchestration and you will be levering it for HA and such and configuring Nova for the same, else if you would just like to use try out openstack on vmware, you can use the ESXi hypervisor directly too and the driver for this is also available in the standard nova installation by default. Hope this clarifies.

Cheers! Venu

edit flag offensive delete link more

answered 2014-06-18 02:40:29 -0600

Shardy gravatar image

updated 2014-06-18 02:41:14 -0600

Thanks for the various answers, it wasn't clear to me from the diagrams if the boxes represented machines (or VMs) or just components running on a machine.

So to sum up in a VMWare environment you need to run a small compute node to send API requests to the VMWare hypervisors (either directly or via vCenter).

Hope I got this right :)

edit flag offensive delete link more


Yes Shardy you are right,

not sure what you wanted to convey with the "small compute node", there should be absolutely no difference to the compute node, as even when you are not using the VMWare as hypervisor, you are using QEMU default inbuilt hypervisor, of-course its with lower footprint :)

Venu Murthy gravatar imageVenu Murthy ( 2014-06-19 00:04:52 -0600 )edit

I guess I'm making a distinction between the machine or VM running the nova/compute software (which doesn't need huge resources) and the machine(s) running ESX hypervisor which do need lots of resources to run VMs... does it make sense?

Shardy gravatar imageShardy ( 2014-06-19 02:01:32 -0600 )edit

answered 2014-06-13 22:45:19 -0600

VMTrooper gravatar image

The Nova implementation requires vCenter to communicate with the vSphere hosts. You will need to have a VM running the Nova Compute service for each cluster that you want to control with OpenStack.

The networking piece is more tricky. If you want to use Neutron, you have to use VMware's NSX product. Otherwise, you will need to use Nova Network with the standard VMware vSwitches.

edit flag offensive delete link more


yes, Except that currently you can manage multiple clusters as compute nodes with one VM which has compute service running.

acbharadwaj gravatar imageacbharadwaj ( 2014-06-16 07:01:04 -0600 )edit

Get to know Ask OpenStack

Resources for moderators

Question Tools

1 follower


Asked: 2014-06-13 20:36:10 -0600

Seen: 547 times

Last updated: Jun 18 '14