Hi Uwe,

The Network Design chapter of the Operations Manual probably comes closest to what you're looking for, AFAIK anyay.

Let's see if I can define some terms for you which is probably half the battle.

A "bridge" is a virtual ethernet bridge that nova-network uses to connect all the virtual machine interfaces on a physical node to the actual network. Under the covers (possibly servel layers of covers) it's the normal linux bridging stuff http://linux.die.net/man/8/brctl is the fine manual page, if you're unfamiliar. You won't need to use brctl directly but seeing the manual may give an idea of what's going on there.

"Fixed" address is the OpenSack way of describing the IP address you virtual instances see if you for instance run 'ifconfig eth0' inside one. The documentation typiclaly shows this as an rfc1918 address that is bridged to a private network on eth1 on your physical nodes. This is not required if you have sufficient public addressing it could use that rather than RCF1918 space. The key is to pick an ip block big enough to handle all the instance you might run at once for many people is requires using private address space. Usually thes address are assigned dynamically though it is possible to specify an address when creating an instance on the commandline (there is not a dashboard webui way to do this)

"Floating" addresses are a bit more difficult to explain. If you are familiar with AWS they are like elastic IPs. These are IPs that are allocated to a project, typically on a public network, and can be dynamically mapped and unmapped to dirrerent running instances. Using the normal operating system tools inside the instance you can't see this address which can be awkward. It's main purpose (as I see it any way) is public connectivity to some instances, for example load balancers or other public facing services. The docs usually show this network on eth0 on the physical hosts, though it could be another interface in a particular setup. OpenStack uses iptables SNAT rules to map the "floating addresses" it assigns as alternate IPs on that interface to the virtual interface of the instance it's currently mapped to.

You haven't asked yet but there are a couple other terms that usually float aroun nova-network deployments:

"Flat DHCP Network" is a configuration where OpenStack maps all the "fixed" network interaces to a single layer2 network, this is commonly used in "multi host HA" mode where each physicla node provides NAT and DHCP service to the VMs running on it. Without the multihost config both NAT and DHCP need to be on a single hoast which becomes both a bottle neck and single point of failure.

It is also possible to set up "Flat Network" which is the same as above but without OpenStack handling DHCP. Using an external DHCP server is possible but then the openstack utilities ... (more)