[c-nsp] IP LFA in ring topology

[Saku Ytti]

On (2012-11-29 13:48 +0100), Adam Vitkovsky wrote:

> What brings us to mpls-te primary and backup tunnel from each node to every
> other node on the ring
> And that is crazy as well

MPLS FRR with auto-tunnels?

There is no technical reason why we couldn't do MPLS FRR without any
RSVP-tunnels, based purely on pre-calculated IGP.

Say you have the topology and fault I wrote earlier. And you have global
IGP labels.
Then when R5 notices the fault, could impose label R3 on top of existing
[P1, VPNn] stack it had. 
So stack would be [R3, P1, VPNn], R5 would push this to R4, which would not
return it to R5, as it's going towards R3, which path is unaffected by
failure.
At R3 the label would be popped, and forwarded based on original [P1, VPNn]
stack, which is also unaffected by the fault, from R3 POV.

All this is fully local repair, any topology, any single fault, before,
after and during IGP reconvergence. R5 has all the information it needs.

Fast and local repair is focus on standard bodies and I'm sure something
like this, 0 config to operators, will appear to the patient.

-- 
  ++ytti