[c-nsp] ASR920 port based xconnect equivalent

[Jason Lixfeld]

> On Jun 7, 2016, at 3:54 PM, Adam Vitkovsky <Adam.Vitkovsky at gamma.co.uk> wrote:
> 
>> Lukas Tribus [mailto:luky-37 at hotmail.com]
>> Sent: Tuesday, June 07, 2016 8:36 PM
>> To: Mark Tinka; Adam Vitkovsky; Erik Sundberg; Stephen Fulton; cisco-
>> nsp at puck.nether.net
>> Subject: AW: [c-nsp] ASR920 port based xconnect equivalent
>> 
>>>> -isn't L2PT required only in case the providers network is actually L2
>> switched environment (running STP internally)?
>>> 
>>> Not only.
>>> It would be required even in MPLS-switched backbones.
>>> Ethernet is Ethernet, regardless of the core transport protocol.
>> 
>> Forwarding of L2 protocols instead of tunneling can be easily done. The L2
>> protocols dst-mac is buried under a number of labels (transport and vc) and
>> another ethernet II header, I don't see why this would be punted to the
>> LSR's CPU in a mpls switched backbone, even if not "tunneled" (where
>> tunneling means dst-mac rewriting).
>> 
>> Not sure if I got your point correctly though?
> 
> My understanding is that in MPLS it really depends on the default behaviour of the edge device, that is whether it tries to interpret incoming PDUs or it doesn't care and forwards or drops them.
> James mentioned that ASR9xx family does try to make sense of incoming PDUs by default (or possibly dropping).
> ASR9K family on the other hand forwards by default.

This seems to be an odd choice for Cisco to switch to service instance based xconnect for port based EoMPLS as opposed to port based xconnect, especially if one now needs to take into account forwardability of L2 control traffic.

My recollection is that in old-style port-based EoMPLS on an ME3600, L2 control traffic is stuffed onto the pseudowire and comes out the other side with the SMAC of the sending client device.  Maybe I’m wrong, but I could have sworn that I saw CDP, LLDP and STP through a port-based pseudowire in the past.  Admittedly, I can’t recall if I had ever tested it specifically.  My core-facing topology is all port-based L3 though.  Maybe that matters?