Summary

  • RFC 3036 distinguished an Address Prefix FEC from a Host Address FEC, even for the same full address.
  • RFC 5036 removed the Host Address form because no implementation used it, while keeping prefixes that can span the full address length.

Two names for one destination

When the IETF published the first Label Distribution Protocol specification in January 2001, MPLS still needed routers to agree on what a label meant. LDP's answer was to associate a Forwarding Equivalence Class, or FEC, with the label-switched path. A FEC described the packet set eligible for that path. The protocol then distributed bindings between those classes and labels.

RFC 3036 defined two address-shaped FEC elements. An Address Prefix could be anywhere from zero bits to the entire address. A Host Address carried a complete address. They were not merely two spellings for the same thing: the specification said a full-length prefix and a Host Address for the same value had different effects. In packet mapping, a matching Host Address could take precedence over a prefix match. The distinction was real in the rules and on the wire: the Prefix element had type 0x02; Host Address had type 0x03.

That extra choice did not survive the next base specification. RFC 5036, published in October 2007 as the revision that advanced LDP to Draft Standard, lists changes from its predecessor. First among them: the Host Address FEC and its references were removed because it was “not used by any implementation.” This is unusually clear evidence in a standards document. The text does not name vendors, count deployments or say why implementers left the element unused. It does say that the feature had no implementation use when the specification was revised.

What disappeared—and what did not

RFC 5036 defines one base FEC element: Address Prefix, still valid from zero bits through a full address. The revision therefore removed a distinct packet-class vocabulary item, not the ability to identify a single address with a full-length prefix. The change should not be inflated into a claim that LDP itself was unused, that host routes vanished, or that every old behavior was interchangeable. The standard documents a narrower edit: one separately encoded FEC type had no implementation use; the general prefix form stayed.

This is a small but revealing kind of standards history. Protocol specifications can accumulate options before the ecosystem has shown which distinctions matter in working systems. A later revision can then draw on implementation experience to retire a distinction without abandoning the larger function it was meant to serve. Here, the record is unusually spare: no post-hoc story about a broken feature, no invented vendor survey, just a sentence connecting removal to non-use.

That restraint matters. The RFC gives no implementation census or measured network share, so the claim cannot be extended to “nobody ever sent the type” or “operators rejected it.” Nor does it tell us why the format was unused. Perhaps implementations simply preferred the prefix element; the document does not say. What it establishes is enough: the standards group revised the base protocol, recorded the specific implementation fact, removed the unused element and preserved full-length prefixes.

In the larger MPLS design, that is not a change to label distribution as a whole. The architecture still groups packets into FECs, and LDP still conveys label bindings. The update record instead shows how the vocabulary around an architecture can be narrowed when one encoded distinction lacks implementation support. History here lies less in a dramatic protocol launch than in the quiet deletion that made the specification fit demonstrated practice more closely.

Sources

The Heng Lu essays are disclosed analytical frames; Lu did not author or endorse either LDP specification.