Open Stack

On 07/16/2013 06:46 PM, Ian Wells wrote:
> On 17 July 2013 00:11, Jay Pipes <jaypipes at gmail.com> wrote:
>> Absolutely, that is what our tools team is now having to do. All I'm saying
>> is that this wasn't necessary in Folsom and wouldn't be necessary if the API
>> didn't force networks to be created with a tenant ID.
>
> What's wrong with a shared network?  It's been a while since I was the
> one doing the network setup steps, but I believe you create, perhaps
> with your admin user, a network with --shared, create a subnet for
> that network, make sure it routes appropriately, and voila! There's
> one network in the system and with no nova boot options (note: nova
> boot doesn't need the subnet to get an address, anyway, though you can
> explictly specify network) your VM will join all visible networks on
> the system - being the one and only shared network.
>
> There are certainly other deployment options, such as the
> one-network-per-tenant model you seem to have in mind, but I think the
> above one mirrors the nova-network style you're used to the closest.

The one-network-per-tenant model is what is used in VLAN networking with 
nova-network and what we're trying to emulate here. That said, due to 
the limitations in the Neutron API I've discussed here (not being able 
to create a network with an assigned tenant ID), we have instead had to 
go the route of having one giant shared network, creating a subnet at 
tenant creation time for the tenant, and relying on net-ns overlapping 
CIDRs.

Best,
-jay