Summary

  • RFC 5212 allows a virtual TE link to be advertised as though a lower-layer LSP had been established, while explicitly saying that the underlying LSP does not yet exist.
  • One traffic-engineering database can make a cross-layer route computable. It cannot by itself prove adjustment capacity, resource commitment, verification, upper-layer installation or customer delivery.
  • The defensible receipt names every transition: advertisement, computation, policy, trigger, lower-layer setup, verification, upper-layer setup, forwarding and observation.

The green link at 14:00

At 14:00, a packet-layer controller computes a 10 Gb/s path. On its screen, two edge nodes are joined by a stable traffic-engineering link. The graph is coherent, bandwidth appears sufficient, and the explicit route crosses the lower optical layer without exposing its internal hops.

The displayed link is virtual. It represents a lower-layer LSP that could be created if an upper-layer request selects it. At 14:01 the request triggers signaling. Meanwhile, another transaction has consumed most of the adjustment capacity inside a hybrid boundary node. Its broad external link advertisement still looks generous, but the particular function that must terminate one switching type and hand traffic to another has only 2.5 Gb/s left. The lower-layer setup fails. At 14:02 the service console still says path available, because that label is showing the earlier computation rather than the later execution.

This is a constructed scene, not a report about an operator, product or incident. The path computation was not fraudulent. It was valid against a snapshot and a modeled possibility. The failure begins when the organization turns that limited result into a resource promise, then into a delivered-service claim, without preserving the steps in between.

What RFC 5212 actually specified

RFC 5212 was published in July 2008 as an Informational IETF document. It describes requirements for GMPLS-based multi-layer networks and, as a special case, multi-region networks that contain different switching technologies. A layer describes switching granularity; a region corresponds to a switching capability such as packet, Layer 2, time division, wavelength or fiber switching.

The document does not define one complete protocol solution. Its scope is a single GMPLS control-plane instance controlling several layers within a traffic-engineering domain. That unification can reduce duplicated routing instances, support rapid provisioning and allow path computation to consider resources across layer boundaries.

Those benefits do not collapse the underlying realities. RFC 5212 says that consolidating identifiers and traffic-engineering information does not alter the separation between control plane and data plane. It also recognizes that two layers can sit under different administrative control. A shared view is therefore neither one physical fabric nor one organization.

A database of choices

In the integrated model, TE links from multiple layers may be consolidated into one Traffic Engineering Database. A path computation process can then choose a route that crosses regions and uses a lower-layer LSP to carry an upper-layer LSP. This is powerful because the computation is no longer blind to the supporting transport system.

But a TED is a database of advertisements. Its contents describe topology, switching capability, encoding, bandwidth and other constraints as reported under particular rules and at a particular time. A computed route is evidence that an algorithm found a candidate through that representation. It is not evidence that every resource on the candidate has been reserved.

That distinction becomes operational at a layer boundary. A hybrid node may need internal termination and adjustment resources to move a signal between switching technologies. A broad external capacity figure can remain high while the specific internal adaptation function is depleted. RFC 5212 therefore requires the adjustment capability to be advertised and expects path computation to treat available adjustment resources as a constraint.

The word available still needs a timestamp, granularity and owner. An advertisement may be current when flooded and stale when selected. Aggregation may hide the resource that matters. A reservation racing with another transaction may lose. The map can be accurate as a map and still fail as a promise.

The link that is deliberately not there

RFC 5212 makes the boundary unusually explicit in its treatment of virtual TE links. Pre-establishing every lower-layer LSP would waste bandwidth and adjustment resources when no upper-layer traffic needs them. The document therefore permits a lower-layer possibility to be advertised upward as though it were a fully established TE link, without actually creating the underlying LSP.

That is not a defect or a deception. It is an intentional option contract. If an upper-layer LSP uses the virtual link, the lower-layer LSP must be signaled immediately. The virtual link lets a stable upper-layer topology coexist with resources created only on demand.

The operational error is to erase the adjective. Virtual TE link selected means that a subsequent transaction is now required. It does not mean bandwidth is committed. Trigger sent means an attempt began. It does not mean the attempt succeeded. Lower-layer LSP established still does not mean its connectivity and data integrity were verified, the upper-layer LSP was installed, or customer traffic crossed it.

A trustworthy console must express those states separately. A single green edge cannot carry them all.

A hierarchy hides both complexity and risk

Once a lower-layer LSP exists, it may be advertised to the upper layer as a TE link or Forwarding Adjacency. Several higher-layer LSPs can be nested over it. The resulting Virtual Network Topology gives the client layer a simpler graph and hides the route taken inside the server layer.

The simplification is valuable, but RFC 5212 identifies its price. Two apparently separate optical paths can share one fiber and fail together. If shared-risk information is not carried across the abstraction boundary, the upper layer may compute diversity that exists only on the diagram.

This is not an argument for exposing every internal detail. Abstraction is essential for scale, commercial separation and administrative autonomy. It is an argument for specifying which facts must survive abstraction: shared-risk groups, effective capacity, protection state, verification status, alarm mapping and the identity of the authority that can commit or release resources.

The same discipline applies to commercial boundaries. A lower-layer provider can expose a constrained service without revealing its topology. The upper layer still needs an executable promise, a correlation identifier and an outcome—not a hopeful interpretation of an advertised shape.

Evaluation did not turn requirements into execution

RFC 5339 later evaluated the then-existing GMPLS protocol suite against the RFC 5212 requirements. Its conclusion is more nuanced than either “GMPLS already did everything” or “GMPLS did nothing.” Many functions depended on local procedures, operator policy, traffic-engineering algorithms and management systems outside the protocol specifications.

The evaluation identified six areas for extensions, including setup and deletion of virtual TE links, graceful TE-link deletion, constrained multi-region signaling, advertisement of adjustment capacity, coordination across per-layer management modules and configuration of technology-specific OAM.

Its adjustment-capacity example is especially instructive. Without link bundling, separate advertisements could expose the constrained adaptation resource. With bundling, the aggregate could make a node look capable even though it could not terminate the requested LSP. The problem was not false arithmetic. It was loss of the decisive internal constraint during abstraction.

RFC 5339 is a 2008 evaluation, not a verdict on every current implementation. Its lasting value is methodological: standards requirements, protocol mechanisms, local algorithms and deployed behavior occupy different evidence layers. A citation to one cannot substitute for a receipt from another.

Verification is an independent gate

RFC 5212 says that a lower-layer LSP may be checked for correct connectivity and data integrity before it is made available. The methods depend on the data-plane technology, and the GMPLS control plane should coordinate them. RFC 5339 likewise points to technology-specific mechanisms and treats cross-layer test coordination as an operational matter.

That wording prevents signaling from becoming an oracle. A successful setup response can show that a control-plane transaction completed. Verification can show whether the intended data link is connected and intact. Upper-layer signaling can then show whether the client path was installed. Probes and counters can show whether traffic used it. Customer observation can show whether the service objective was met.

Each result answers a different question. Combining them into one state removes the exact boundary needed during failure. If a customer sees loss, operators must know whether the candidate was stale, the trigger failed, the lower layer was misconnected, the upper-layer cross-connect was absent, OAM alarms were not mapped, or the data plane diverged after installation.

The receipt worth keeping

Start with the request: endpoints, bandwidth, priority, switching types, protection and service objective. Freeze the TED snapshot used by computation, including its age, source advertisements, filters and missing information. Record the explicit route and every layer or region transition.

For each abstract hop, say whether it is physical, an established FA-LSP, a real TE link or a virtual TE link. At each hybrid node, record the required adjustment function, its capacity and reservation epoch. Preserve the policy decision and management approval that allowed a trigger.

Then follow execution. Link the trigger to the lower-layer signaling transaction; retain the selected server-layer path, committed resources, installed cross-connects and advertisement propagation. Record the connectivity and integrity test before declaring the lower link ready. Preserve upper-layer installation, OAM mapping, shared-risk state, probes, counters and customer observation.

Finally, keep teardown evidence. A released resource with a stale advertisement is the inverse of a virtual link: the map says something exists after execution has removed it. The receipt should name the remaining state as plainly as the successful one.

Sources