On Wed, Mar 13, 2013 at 01:09:16AM +0200, Itamar Heim wrote:
On 03/11/2013 05:16 AM, Mark Wu wrote:
>On 03/08/2013 05:16 AM, Dan Kenigsberg wrote:
>>On Thu, Mar 07, 2013 at 03:57:49PM +0100, Adrian Gibanel wrote:
>>>Just in case it might help you please check:
>>This is almost 1 year old, but I did not notice it yet. I love the
>+1 on NAT network. Except that it can save ip address, it also could
>reduce the external physical switch's pressure on mac table. Because the
>mac address is invisible to external switch.
>But there're two limitations of NAT network compared with physically
>1. The VMs attached to the same NAT network, but on different hosts
>can't hear each other. It could be resolved by constructing a tunnel or
> among the hosts in the same cluster and centralizing the mac
>address management of dnsmasq on ovirt engine.
>2. The VMs in NAT network are hidden behind the host. The external host
>can't initiate a connection to the VM. I think it's fine for a desktop VM.\
>For a server VM, it can't be resolved by add a DNAT rule on demand. It's
>similar to the 'floating ip address' in quantum.
also need to remember live migration will probably not work with NAT.
That's why I consider this as an option to host-local networks.
how would floating IP work? wouldn't you need to map it 1:1 with
>>Yes, the rant there, about ovirt network being tightly-coupled with a
>>physical interface, is 100% justified. I'm trying to address some of
but it's a long
>>way to go.