Re: [mpls] Last Call: draft-ietf-mpls-ldp-upstream (MPLS Upstream Label Assignment for LDP) to Proposed Standard

[Eric Rosen <[e-mail]>]

With regard to this draft, I need to reiterate a comment I made during WG
last call, as I think there is a procedural issue that needs to be brought
to the IESG's attention.

The draft has a normative reference to RFC 3472 "Generalized Multi-Protocol
Label Switching (GMPLS) Signaling Constraint-based Routed Label Distribution
Protocol (CR-LDP) Extensions", despite the fact that CR-LDP has been
deprecated by RFC 3468 ("The MPLS Working Group decision on MPLS signaling
protocols", February 2003).  I don't think this is allowable; I interpret
RFC 3468 as requiring that we do not take any action that prevents the
deprecated CR-LDP documents from being reclassified as historic.

The only reason the CR-LDP documents were not classified as historic seven
years ago is that there are other standards organizations that have produced
specs with normative references to CR-LDP.

Section 4.3 of RFC 3468 says:

   standards organizations which reference the document [the CR-LDP specs],
   need to be notified of our decision so that they (at their own pace) can
   change their references to more appropriate documents.  It is also
   expected that they will notify us when they no longer have a need to
   normative reference to CR-LDP.

I think the clear implication of this is that neither the IETF nor any other
SDO should be creating new normative references to CR-LDP documents. Note
that RFC 3468 explicitly calls out RFC 3472 as one of the deprecated
documents.

During WG LC, one of the authors stated that he disagreed with my
interpretation, but WG consensus on this issue was never obtained.

At this point, I believe the only change that is needed to draft-ietf-mpls-
ldp-upstream is to move the reference to RFC 3472 into the "Informational
References" section.  That is, I think that recent revisions to the draft
have made the normative reference gratuitous, as one does not need to read
RFC3472 in order to implement any part of this draft.

If the draft is published with this normative reference, it is almost
inevitable that someone will someday say "we don't want to use LDP
upstream-assigned labels, because they depend on CR-LDP and CR-LDP has been
deprecated".