Hi All

I am experimenting with a system comprising Cisco UCS Blades and Nexus
5k switches.

I am trying to understand the best way to define network interfaces for
my vms, without having to specify exact source device names or PCI
addresses, as I want to be able to migrate vms between nodes and not
have to worry about namespace or pcispace collisions.

At the moment my network definition looks like:
<interface type='direct'>
<mac address='52:54:00:c0:cb:2a'/>
<source dev='eth55' mode='private'/>
<virtualport type='802.1Qbh'>
<parameters profileid='TestKVMProfile'/>
</virtualport>
<model type='virtio'/>
<driver name='vhost'/>
</interface>

libvirt added a few extra settings when I added it:
<target dev='macvtap0'/>
<alias name='net0'/>
<address type='pci' domain='0x0000' bus='0x00' slot='0x03'
function='0x0'/>

The issue I have with this config is that I am specifying a source dev
(eth55), which I would have to keep manual track of across all vms in my
infrastructure.

Is there a way to use a high-speed virtual network interface, which is
created on the fly for the vm, and supports migration cleanly to another
kvm host (ie, does not rely on any specific host resource names).

Sorry for any confusion caused from my lack of understanding :)

Thanks
Chris