Mon Feb 20, 2006 10:18 pm
Q in Q does not have anything directly to do with VPNs. It is simply the layering of one VLAN tag upon another VLAN tag. Just basic double encapsulation.
It's a very useful techneique, but only works where you maintain a layer 2 infrastructure. The most common uses for Q in Q usually involve multi-access common carrier networks, where multiple ISPs can share the same physical network, be completely isolated in their own VLAN, yet not lose the ability to do their own VLAN scheme within their virtual lan.
This use is similar to some of the more common uses of non-VPN MPLS, except that it requires a consistant underlying layer 2 infrastructure, while MPLS operates at layer 3 (well, mostly...).
For your application, neither Q in Q, nor MPLS fits the bill exactly, athough you may find either or both useful (VPLS would be nice in this case).
MT has not, as of yet, implemented any of these. Apparently Q in Q will be first.
Here is an idea for your setup:
Setup your radius servers to pass back a framed-ip-address, as well as an in-filter and an out-filter.
Keep your various VPN groups tied to a particular range of IPs, and setup your in-filter and out-filter chains to allow communication only between addresses within the same range.
Then, depending on the scale of your opperation, set something up to properly attract the traffic to the destination concentrator. For a very small setup, proxy-arp would be fine; with a somewhat larger setup (not huge), OSPF with redistribute-connected would do the trick. With a huge setup, it becomes more complex.
--Eric