[c-nsp] EoMPLS from ME3750 to 7201 GigE sub-int

[Justin Shore]

Justin Shore wrote:
> I'm having trouble getting EoMPLS to work from a ME3750 to a 7201.  I 
> have some signaling but the VC just won't come up.  The basic layout is 
> a switchport on the ME3750; the xconnect is in the SVI associated with 
> that switchport on the ME3750.  The ME3750's P-facing interfaces are 
> MPLS enabled.  The 7600 core is MPLS enabled.  The 7201 dual-homes to 
> both core 7600s.  The 7201's interfaces are MPLS enabled.  I'm trying to 
> terminate the xconnect on a sub-int of an onboard interface that's 
> facing a 4948.  I have a laptop on the ME3750 and I'm mapping an access 
> int on the 4948 in the correct VLAN back across another network to my 
> office (so I don't have to sit in a noisy head-end).
 >
> Something just jumped out at me.  The MTUs are different.  Would that 
> cause an issue?  Gi0/2 on the 7201 is set to 9000 because hung off of 
> the 4948 is a set of backhaul radios that I will be doing MPLS across. 
> The customer-facing interface on the ME3750 is set to the default 1500. 
>  Do the MTUs have to match up?  I'm going to have to do some research.

The MTU thing got me thinking (dangerous I know).  I set the MTU on 
Gi0/2 on 7201 back to 1500 and the VC immediately came up and passed 
traffic.  The MTU on the sub-int of the 7201 can't be set separately 
from the physical interface's MTU:

% Non-TRISL encapsulated sub-interface GigabitEthernet0/2.130 does not 
support user settable mtu.

MPLS MTU on the sub-int or physical interface doesn't appear to affect 
EoMPLS VCs either.  Is there a way to change the MTU on the sub-int to 
allow the VC to come up?

Along this same train of thought, what's best practice for EoMPLS links 
when mixed with sub-ints?  Should I strive to keep infrastructure links 
separate from customer links?  Ie don't put them together on the same 
interface even if separate by sub-ints?  In this example Gi0/0 and 1 on 
the 7201 are what dual-home the 7201 to the 7600 core.  Gi0/2 connects 
to the 4948.  Gi0/3 is currently unused.  I could also connect Gi0/3 to 
the 4948.  In that case I could then dedicate that interface to the 
infrastructure link (only the radio at this point) on the 4948.  Since 
I'm standardizing on a MTU of 9000 for infrastructure links I could keep 
them all on the same physical interface of the 7201.  For that matter in 
the short-term I could also just connect the radio directly to the 7201.

So what's best practice for EoMPLS on sub-ints?  Had I not run into this 
problem this afternoon I'd have run into it next week during another new 
deployment.  In that case I'll be on a 7206 with only 3 GigE ints but I 
could make it work.  Thoughts?

Thanks
  Justin