Ask Your Question

Convert simple KVM network to Openstack

asked 2014-09-20 09:21:44 -0600

Anonymouslemming gravatar image

updated 2014-09-20 14:11:46 -0600

larsks gravatar image

I'm trying to convert my simple KVM based home lab to Openstack, and I'm struggling to find instructions for how to setup the network.

My current setup has 2 hypervisor hosts running Ubuntu 14.04 LTS and KVM. Each host has 2 NICs, but I'm currently only using one. Each NIC is setup as a bridge (br0) with a static IP address. Guests running on each hosts have IP addresses from the same range. The default gateway for all hosts on my network is a PFSense firewall with an internal address of The external address is allocated by the ISP and I NAT my private network outgoing connections to that.

Existing Setup Diagram - A diagram can be seen at

image description

I want to move to Openstack for learning purposes, and I'm a bit lost as to what components I need to use and how to configure them. The icehouse neutron documentation at is more complex than what I'm looking for.

I can find roughly what I think I'd want in the Havana docs at but I can't work out how to translate that into a working example.

Proposed Setup - What I'd like to get to is diagrammed at image description

I want to use a single host for the network and controller node. The controller will run as a guest on the network node. The network node and compute nodes have 2 NICs. Each NIC is connected to a different switch (I don't have VLAN enabled switches yet). There is a management network on and the compute node, the network node and the controller node can all reach each other on this.

I've done the controller setup steps up to and including the compute service setup ( ). Where do I go from here ?

Thanks in advance,

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted

answered 2014-09-21 21:25:03 -0600

larsks gravatar image

I would stay, "start simple."

Rather than trying to make the controller a virtual machine on your network node, why not just colocate the "controller" and "network node" services on the same piece of hardware without virtualization? This is a very typical and fairly well documented configuration, and is pretty much what you get "out of the box" using tools like packstack with dedicated compute nodes. I think this will probably simplify the networking aspect of the deployment.

You don't need VLANs enabled in your environment, provided that you elect to use Neutron as your network service. This supports GRE and VXLAN overlay networks, which give you network segmentation without needing to coordinate the configuration of your physical switching infrastructure.

You controller will need one "internal" interface, over which it will communicate with your compute nodes, and one "external" interface, over which you will offer "public" API services and provide external addresses for your Nova instances.

You may find it helpful to stop by the #openstack irc channel and work through some of your questions in a more interactive fashion.

edit flag offensive delete link more


One reason I've been building the controller in a virt is that it makes it easy to blow it away and rebuild it when required, but I'll rebuild shortly as that makes sense.

Do you have any docs that show how to configure Neutron with a flat network so that nova instances get IPs from the ext range ?

Anonymouslemming gravatar imageAnonymouslemming ( 2014-09-22 14:07:28 -0600 )edit

Get to know Ask OpenStack

Resources for moderators

Question Tools

1 follower


Asked: 2014-09-20 09:21:44 -0600

Seen: 582 times

Last updated: Sep 21 '14