The MPLS WG Archive

[Philip Matthews <pmatthews@hyperchip.com>]

Title: Drafts on LDP Resiliency

For over a year now, the MPLS WG has been working on
a draft (draft-ietf-mpls-ldp-ft-02.txt) which describes
how to keep LSPs set up using LDP alive across control
plane restarts and activity switches.

At the recent MPLS WG meeting in Salt Lake City,
two new drafts were presented that proposed new approaches
to this subject.

The groups behind these three drafts met after the meeting
with George Swallow (co-chair), Scott Bradner and Bert Wijnen
(Area Directors) to discuss the pros and cons of the various
proposals and how best to move forward.

For pros and cons, we came up with the following list:

* draft-smith-mpls-ldp-restart-00.txt
  Aimed at edge routers doing VPNs and L2 Circuit
  Emulation where all LSPs originate or terminate on the box,
  and LSPs are pretty static.
  Pros:
      - Only a very small amount of re-signaling after a restart
        or switchover.
  Cons:
      - Must save LDP state across restart/switchover.
      - LSPs after last checkpoint are lost and need re-signaling.

* draft-leelanivas-ldp-restart-01.txt
  Aimed at routers which cannot save LDP state across a restart/switchover.
  Pros:
      - Works for transit lsps, as well as originating/terminating lsps.
  Cons:
      - Must re-signal all lsps. (People are divided as to how bad this
        really is, but it does take time and effort).

* draft-ietf-mpls-ldp-ft-02.txt
  Aimes to give good switchover performance.
  Pros:
      - Little signaling after a switchover.
      - Works for transit and orig/term lsps.
  Cons:
      - Must save LDP state across a restart/switchover.
      - May be harder to implement than the Smith draft when starting
        from an existing non-FT LDP implementation.

We agreed on the following approach for moving forward:

* The draft-smith-mpls-ldp-restart-00.txt draft will be
  combined with the current WG doc (draft-ietf-mpls-ldp-ft-02.txt).
  At a minimum, the Smith draft will be added as an alternative
  approach using the same basic technique, but we will attempt
  to integrate the two approaches as much as possible.
  Adrian Farrel has some ideas for this.

* The draft-leelanivas-ldp-restart-01.txt draft will continue
  as a separate draft for now, since the technique is uses is
  quite different from the other two.

* Both drafts will be enhanced with applicability statements
  that clearly states when each approach is appropriate.

- Philip