[c-nsp] dot1q tunnelling presentations

[sthaug at nethelp.no]

> I have a question about cisco's dot1q tunnelling implementation, that the 
> documentation doesn't seem make immediately clear.
> 
> Is it possible to reverse the direction of the tunnelling?
> 
> For example?
> 
> Switch C-A ---- Switch P-A ++++++ Switch P-B ---- Switch C-B
> 
> The (P)rovider is giving the (C)ustomer a P2P link with a single VLAN for 
> their use. The link between C-A and P-A, and P-B C-B is a dot1q trunk, 
> that carries a single vlan, call it vlan X.
> 
> Can Switch C-A and C-B tunnel VLANs over the providers single VLAN. ie. 
> The tunnelling takes place on the client's switches and not the providers.

Yes, 3550 QinQ, and now also 3750 QinQ, can do this. *But* - and this is
a big but: QinQ means that you will have two (or more) levels of VLAN
tagging between provider switches P-A and P-B. Not all types of switches
can handle this, even if Cisco 3550, 3750 etc. can do it.

> It seems to me that it's simply a case of where the 4byte VLAN id shim is 
> added? I'd like it to be added on the Customer switches in the above 
> diagram, rather than the providers.
> 
> Something like:
> 
> switchport mode dot1q-tunnel
> switchport mode access vlan x
> 
> on C-A and C-B would achieve this, if the ingress and egress notion of the 
> switchports could be reversed? (So it's spits out tunnelled VLANs, rather 
> tunnels what it sallows).

Take whatever trunks you need *in* to C-A and C-B as dot1q-tunnel ports,
and push them to the provider as normal trunks. Simple as that.

Steinar Haug, Nethelp consulting, sthaug at nethelp.no