[j-nsp] MPLS fast convergence without link information

[Mark Smith]

Hi list

Let's examine an MPLS network where the connections between MPLS
routers (P and PE) involve L2 devices. The connections might be leased
L2 pseudowires or something similar. As a consequence the interface
link state information cannot be used by routers to determine the
state of the connection or neighboring router. For example consider
routers A and B. If the connection between them is down, both still
see interface (fiber or copper) link up. Similarly in case of failure
of router A, B still sees interface link up. For the sake of
simplicity, let's assume all routers are Juniper (e.g. MX series),
IS-IS is used as IGP and RSVP(-TE) as the MPLS signaling protocol.

We would like to achieve the fastest convergence times safely available.

What is the current best practise to accomplish this and what order of
magnitude are the convergence times?

I have spent some time searching for a solution. I have found these:
1) enable BFD on the IGP session
2) enable BFD on the LSPs
3) enable BFD on the RSVP session

Which one of them (if any) is the best solution?

I also read that BFD might cause IGP to reconverge instead of
switching the traffic to bypass LSP (when using link-node protection).

Any help and pointers to good material is greatly appreciated.

Thanks