[c-nsp] Best Practices for quickly removing routes when BGP peer drops

[Saku Ytti]

On Thu, 11 Dec 2025 at 00:28, Lukas Tribus via cisco-nsp
<cisco-nsp at puck.nether.net> wrote:


> Agreed and based on your other mail I now got your point about XR. Thanks
> for clarifying.
>
> I'm interested to hear why folks don't likel labeling internet traffic.
>
> I always l liked it, indeed I have a real distaste for programming the full
> table to the FIB on core only boxes.
>
> Perhaps thats because I used real slow FIB convergence boxes for a long
> time.
>
> I also like to make a routing decison ideally only once, not even on the
> egress box if possible (per next hop label allocation).
>
> But those are "personal tastes".

Oh I've drank the MPLS koolaid when it was still tag-switching. So I
definitely think labels are best practice.

However some reasons people might not want to do BGP free core is that
they don't know how to get traceroute working without inflated
latencies, which might cause customer tickets. There is an easy
solution to it, but it may not be obvious.

And definitely in addition to labeling in core, I think BGP-LU to have
per-CE labels also to avoid egressPE lookup. Then we could imagine a
network where P and PE core side interfaces have trivial size on-chip
memory for labels, and only edge facing devices can do LPM lookup.
Unfortunately there wouldn't really be economic advantages, as cost is
not in the hardware.
-- 
  ++ytti