Summary

  • RFC 3471 generalized the MPLS control plane so a label could coordinate packet, time-slot, wavelength and fiber switching, even when the forwarding device never inspected a packet header.
  • The generalized label was intentionally contextual: a proposal, allowed set, signaling success or healthy control channel did not by itself prove that the intended physical resource was allocated or carrying traffic.

The word label once evoked a compact value attached to traffic. RFC 3471 kept the word and changed its physical reach. A label could become a coordinate for a repeating time slot, a color of light, a waveband, or a whole fiber port. The conceptual move mattered because it let one control architecture span routers, digital cross-connects and optical fabrics without pretending that their forwarding planes worked alike.

Published on the Standards Track in January 2003, RFC 3471 was a functional description. It did not define every wire format or every optical technology. CR-LDP and RSVP-TE details went to RFC 3472 and RFC 3473; technology profiles went elsewhere. Its job was to state which distinctions a generalized signaling system had to preserve.

It began with four interface classes. Packet-Switch Capable interfaces could act on packet or cell headers. TDM interfaces selected positions in a repeating cycle. Lambda-Switch Capable interfaces selected wavelengths. Fiber-Switch Capable interfaces selected physical spatial paths. These were properties of interfaces, not mystical properties of whole nodes. One device could sit between different switching realms.

That hierarchy could nest. Packet LSPs could ride inside a TDM LSP, which could ride inside a wavelength LSP, which could in turn ride through a fiber-level LSP. Aggregation reduced state, but it also created an evidence problem: observing the outer optical construct did not enumerate every client flow, while observing an inner packet path did not prove the outer resource remained intact.

RFC 3471 therefore separated three fields in the Generalized Label Request. LSP Encoding Type described the nature of the requested signal. Switching Type described what a particular link should do. G-PID identified the client payload, mainly for endpoints. They were related coordinates, not synonyms. “Ethernet carried by a lambda over a fiber” contains at least three layers of fact.

The generalized label itself contained no type field. Adjacent nodes were expected to know, from the interface context, which label format applied. That omission was economical and deliberate, but it made provenance essential. A stored integer or byte sequence detached from local and remote interface identities, switching capability and technology is not a portable account of what resource existed.

Physical resources also came in discrete, scarce units. A link might advertise several encodings while having no free slot or wavelength when a request arrived. Capability meant the equipment could support a form under some conditions; it was not an allocation receipt.

The Suggested Label exposed another boundary. An upstream node could propose a resource and begin configuring slow hardware early. An optical fabric with moving elements might benefit from that head start. Yet the downstream node retained the right to override the suggestion. A proposal could reduce latency, but only the final choice and completed programming established what was actually selected.

A Label Set narrowed the downstream choice. It could express available wavelengths, continuity constraints or equipment limitations. Membership in that set meant “permitted at this stage,” not “reserved,” “installed,” or “working.” As constraints moved hop by hop, an empty intersection could still defeat the path.

Bidirectional LSP setup brought two directions into one coordinated transaction. This avoided races in which two independent setups competed for reciprocal resources. RFC 3471 even specified a tie-break based on source node identifiers for simultaneous attempts. The rule made control behavior deterministic; it did not assign commercial priority, guarantee symmetrical signal quality or show that either endpoint received its payload.

The memo also formalized separation between control and data channels. Optical equipment might exchange signaling on an IP path that was physically distinct from the circuit being established. That made control possible where in-band packets did not exist. It also broke a convenient inference: a live control adjacency did not prove a live data path, and a broken control channel did not necessarily mean that an already programmed optical path had stopped carrying light.

Protection and administrative-status fields were likewise statements of intent and management state. Requested protection was not proof of diverse capacity. “Testing,” “up,” or “delete” in a control object was not a photonic measurement. Authentication could establish who sent a signaling message without attesting what the cross-connect did afterward.

The later GMPLS architecture, routing requirements, OSPF extensions, link bundling, LMP and G.709 work made those boundaries more concrete. They did not retroactively prove a universal deployment. The history is one of decomposition: a common coordination vocabulary at the center, with technology-specific facts and operational receipts around it.

Heng Lu's running-code principle makes the lesson sharp. The label is a symbolic instruction until hardware and traffic evidence show an effect. Minimum specification explains the modesty of RFC 3471's common layer: define enough coordinates to interoperate, while leaving physical allocation and technology detail where they belong. Reality layers prevent a capable interface, allowed label, accepted request, programmed switch, observed signal and delivered application from collapsing into one green status.

An honest GMPLS record therefore keeps a chain. Name the interfaces and their switching contexts. Preserve the advertised capability and current constraints. Record who suggested a label, who selected it, what was programmed in each direction, which control channel carried the exchange, what physical signal was observed and what payload reached the service boundary. RFC 3471's achievement was not to make photons behave like packets. It was to let one control vocabulary coordinate both without confusing the vocabulary for the world it described.

Sources