Simple OVN setup in 5 minutes

Open Virtual Network (OVN) is an awesome open source project which adds virtual network abstractions to Open vSwitch such as L2 and L3 overlays as well as managing connectivity to physical networks.

OVN has been integrated with OpenStack through networking-ovn which implements a Neutron ML2 driver to realize network resources such as networks, subnets, ports or routers. However, if you don’t want to go through the process of deploying OpenStack, this post provides a quick tutorial to get you started with OVN. (If you feel like it, you can use Packstack to deploy OpenStack RDO which by default will use OVN as the networking backend).

After this, we have a functional OVN system which we can interact with by using the ovn-nbctl tool.

As an example, let’s create a very simple topology consisting on one Logical Switch and attach two Logical Ports to it:

What now? Can we communicate vm1 and vm2 somehow?
At this point, we just defined our topology from a logical point of view but those ports are not bound to any hypervisor (Chassis in OVN terminology).

For simplicity, let’s bind both ports “vm1” and “vm2” to our chassis simulating that we’re booting two virtual machines. If we would use libvirt or virtualbox to spawn the VMs, their integration with OVS would add the VIF ID to the external_ids:iface-id on the OVN bridge. Check this out for more information.

At this point we have both vm1 and vm2 bound to our chassis. This means that OVN installed the necessary flows in the OVS bridge for them. In order to test this communication, let’s create a namespace for each port and configure a network interface with the assigned IP and MAC addresses:

After this, we should be able to communicate vm1 and vm2 via the OVN Logical Switch:

In essence, what this post is describing is the OVN sandbox that comes out of the box with the OVS source code and presented by Russell Bryant . However, the idea behind this tutorial is to serve as a base to setup a simple OVN system as a playground or debugging environment without the burden of deploying OpenStack or any other CMS. In coming articles, we’ll extend the deployment to have more than one node and do more advanced stuff.