[c-nsp] Sub-int based EoMPLS on a 6700 series LC in a 7600

[Justin Shore]

I have a TAC case open right now regarding a new EoMPLS VC I'm trying to 
turn up.  This is the first I tried to terminate in a sub-int on one of 
my 7600s.  The LC is a 6748 w/ DFC.  The sup is a 720-3BXL running SRB1. 
  The other end is a ME3750 running 12.2(50)SE.  The ME's end is a 1Q 
trunk facing the CE.  The native VLAN is for Internet back to a SVI. 
The other VLAN allowed on the trunk (tagged) has a corresponding SVI 
with the xconnect to the loopback on that particular 7613.  The 7613 has 
a 1Q trunk facing the CE as well.  The port is configured with a sub-int 
though and the xconnect sits in the sub-int with 1Q encapsulation 
(tagged).  The VLAN IDs are the same on both ends.  MTU is the same on 
both ends and is the default 1500; I'm thinking that I need to raise 
this for one thing, to support the 1Q trunk over EoMPLS.  Or does MPLS 
auto-adjust for the higher MTU?

The VC is up end to end.  I can do a MPLS ping from end to end.  When I 
try to ping from the CE on the ME3750's end 2811 (with Fa0/0 configured 
with sub-ints and a 1Q config matching the ME) I get nothing back and no 
ARP entry on the 3560E.  I do however see counters on the VC increment 
end to end.  When I try to ping from the CE on the 7613's end (a 3560E 
with a trunk configured for the 1 EoMPLS VLAN and its L3 interface in a 
SVI) to the 2811 I get nothing back but the 2811 gets a matching ARP 
entry for the 3560E.  The counters on the VC do not increment either 
like they should (ping with a timeout of 0 repeating 10k times).  That's 
the exact opposite of what I would expect.

One question that was raised is if the core-facing interface had to be 
fancy WAN interfaces.  My understanding is that they do not because I 
don't have to double-tag anything.  The core-facing interface is on the 
same 6748.  For grins I moved the CE-facing interface to a 6724 w/ DFC 
in the same chassis.  No change.  The VC is up but I only get an ARP on 
the 2811 and nothing back.  From the perspective of the CEs traffic only 
flows from the 3560E to the 2811.  From the perspective of the VC's 
counters it's just the opposite.

The 7613 connects to a ME6524 (MTU 9000 on a L3 sub-int on both ends) 
and that ME6524 connect to the ME3750 (also a L3 interface with a MTU of 
9000).  MPLS is enabled end to end.  I get the targeted LDP on both ends 
of course.  IS-IS L2 is my IGP.

Do I have to have WAN ints for the core-facing links?  I'm thinking no 
but I can't say for certain.  Neither could my TAC engineer but they are 
double checking.  Anyone know?

Thanks
  Justin