[c-nsp] Unified MPLS - Discrete area or separate IGP in Access Layer

[Adam Vitkovsky]

> But then again, because it's BGP, you should be able to apply 
> filtering to limit what gets into the FIB, so yes, it could work.

Right using RFC3107 to stitch separate levels/areas/processes will allow us
to introduce routes selectively (via communities). 
The only things that really need to be injected into the access are just
couple of subnets hosting the Packet Core. 
And possibly some PW head/tailends or loopbacks of PEs that need to
participate in L3VPN. 

adam
-----Original Message-----
From: Mark Tinka [mailto:mark.tinka at seacom.mu] 
Sent: Thursday, February 06, 2014 5:23 PM
To: Adam Vitkovsky
Cc: cisco-nsp at puck.nether.net
Subject: Re: [c-nsp] Unified MPLS - Discrete area or separate IGP in Access
Layer

On Thursday, February 06, 2014 05:25:49 PM Adam Vitkovsky
wrote:

> And I did support this. However now facing real world restrictions of 
> 12K routes per ME (app template) or
> ASR901 I could probably use some separation for future proofing the 
> network.

Practically, when you run L1/L2 levels, you end up having to do Route
Leaking (i.e., redistribute L2 into L1), meaning you come off with the same
number of routes as you had before going multi-level.

You could summarize L2 routes and redistribute those into L1, but you end up
with sub-optimal routing (which is why I recommend turning off the ATT bit
when doing L1/L2 routing anyway).

> Way back when I haven't had any major problems with inter-area or 
> inter-as MPLS-TE, or MVPN for that matter -but that wasn't on ME :).

You can use expanded loose hops to string an LSP across IGP level domains,
but it adds complexity.

Also, p2mp RSVP-TE didn't support expanded loose hops until now (well, there
is an I-D that I saw going around).

> So ISIS L2 everywhere though Edge and Metro-E Access rings are not 
> continuous L2 but rather separate domains interconnected with BGP-LS 
> please?
> 
> -Which I would now accomplish with RFC3107 in the meantime.

Right, although I'm not sure BGP-LS will buy you FIB space savings, as the
idea is, really, to redistribute IGP routing across a BGP domain. So you get
scaling due to avoiding IGP- style messages, but not sure you save much on
actual routes.

But then again, because it's BGP, you should be able to apply filtering to
limit what gets into the FIB, so yes, it could work.

Mark.