Summary

  • A 2 November individual Internet-Draft proposes area-scoped distribution for BGP-LS-SPF: detailed Node and internal Link NLRIs stay local, while Prefix reachability can cross an area boundary.
  • Boundary routers translate a remote destination into a locally reachable exit. An area path can reject an advertisement that would revisit an area; the draft explicitly leaves packet-loop freedom with multiple exits, alternate paths and independent metrics to later route-selection rules.
  • The decisive next specification is the representation of a remote prefix, its metric and competing boundaries, followed by a forwarding test across two exits and a boundary failure. Revision 00 supplies no such rule or measurement.

Give Area C a destination P, connect it through B to A, and ask router A1 to send a packet. In the draft's three-area example, A1 does not learn the individual links around P in C. It learns that P can be reached through A's boundary toward B. A1 runs SPF on its own map to reach that boundary. The next area makes its own decision. Detailed remote topology has stopped at the wall; the promise of a route has crossed it.

That is the proposed economy. RFC 9815 defines the base BGP-LS-SPF model using link-state information and distributed SPF. In a large, frequently changing topology, sending every detailed change everywhere makes distant routers carry information that may have no bearing on their next hop. The new draft gives each router an Area ID, identifies a boundary router by an adjacency into another area, and scopes Node NLRIs and internal Link NLRIs to their own area. Links crossing an area may be visible to the neighboring areas. Prefix NLRIs are the goods allowed through customs.

The distinction matters during a failure. The draft imagines an internal B1–B2 link in B failing while the boundaries to A and C remain usable. B handles the changed Link NLRI and recomputes its local path. If the externally advertised reachability and relevant attributes for P stay the same, A need not receive B's detailed link change or repeat SPF because of it. If the B–C boundary link fails with no other crossing, B must update or withdraw P's inter-area reachability. The trigger is the effect on the exported promise, not the mere fact that a local link moved.

Now give B two exits. The proposal says a boundary associates a received prefix with a locally reachable boundary so that local SPF can get traffic to it. It does not yet define the exact representation of that prefix, how metrics from separate areas combine, or how multiple candidate boundaries are chosen. Those are not decorative encodings. They decide whether A and B agree about which exit actually carries P after a change. The draft also defers exact Area ID signaling, area-path encoding, mixed deployment behavior and the interaction of scoped distribution with RFC 9815 origination, sequence numbers and route reflectors.

The loop rule exposes the divide cleanly. If P's advertisement travels C→B→A, a later attempt to advertise it back into B is rejected because B already appears in the area path. That prevents the control message from touring the same area twice. A packet does not inspect that editorial history. With multiple boundaries or independently calculated metrics, each router needs a forwarding choice whose consequences are consistent across areas.

The draft itself states that its area-path check addresses control-information propagation loops while loop-free packet forwarding additionally depends on route-selection and reachability-representation procedures still to be specified.

There is a security owner behind the abstraction. A compromised boundary router could suppress local topology, invent remote reachability or tamper with area-path information, as the draft's security section says. Another area cannot inspect the detailed remote map that the boundary deliberately withheld. The boundary therefore needs both authority to summarize and evidence that its summary still leads somewhere. The draft gives a conceptual architecture, not a deployed trace or measured convergence result. Its own example of fast-changing low-Earth-orbit satellite networks is a proposed application, not a customer reference.

The next useful revision would make the bargain executable: define what a remote-prefix advertisement contains, how competing exits and metrics are compared, when an altered boundary withdraws or replaces a claim, and what packets do through a two-exit failure. A captured inter-area advertisement sequence beside FIB and packet traces would show whether the area path and the forwarding path agree. Until those facts exist, the boundary can promise that the map stays private. The packet will ask whether the promise was worth anything.

Area-Scoped Link-State Distribution for BGP-LS-SPF, revision 00 · RFC 9815