[j-nsp] Stiching L2 to L3 on MX480

[Chris Kawchuk]

Correct.

With the LT- method you'd have the same problem. You'd have to stitch every L2CKT to an LT-0/0/0.x then from LT-0/0/0.y (it's partner) to the L3VPN; so lots of config there too. Likewise, LT- interfaces are bandwidth-constrained (as they use PFE bandwidth to send the packet through the Trio "twice" so to speak; it's equivalent to using a hair-pin cable between to physical ports on the MX; hence could be congested). MPLS-LDP Mesh groups don't pass throughout twice (watch Doug H or ytti correct me here hah)

If you want to keep it simple config on the MX -- you're almost better off "book-ending" the L2circuit with another ACX sitting right next to your MX (i.e. do all the Martini stuff between ACX-in-the-field to ACX-at-the-head-end-POP), and hand off all the L2 circuits to the MX on a physical 10G port. Single place to provision the L3VPN interfaces then on the MX; use per-unit scheduler to enforce southbound QoS at the MX. All the L2 stuff stays on the ACX, all the L3 stays on the MX.

Downside is you'll lose MX-termination redundancy now though and have an extra box at your head-end; which is one thing the ACX-to-L2-to-MX-VPLS solution solves. (the ACX you can terminate the same L2CKT on 2 different MXes PE's using backup-neighbour <alternate-PE-loopback> statement on the ACX at the pickup point "in the field"). 

Effectively this is just Hierarchal VPLS I'm proposing. Have POC'ed this before and worked well; but yes, 1 VPLS = 1 Tail + 1 IRB into the VRF

- CK.


On 20 Dec 2017, at 9:53 am, Pshem Kowalczyk <pshem.k at gmail.com> wrote:

> Hi,
> 
> If I did it this way, wouldn't I need a separate VPLS for every access circuit on the ACX? I'm trying to limit this to one 'infrastructure' L2 setup, with all of the provisioning on MX.
> 
> kind regards
> Pshem