Summary

  • RFC 1917 appealed to holders and providers to return unused IPv4 assignments, but encouragement was not a revocation order and “unused” did not have one observable meaning.
  • A missing global route could coexist with a working private network, while a registry change could coexist with stale configurations and external dependencies. Assignment, visibility, need and withdrawal required different evidence.
  • Later registry architecture made the separation clearer: registration protects uniqueness, operators control advertisement, and present recovered-space records do not preserve the complete cause or history of every return.

A form could change the record, not the world around it

RFC 1917 was BCP 4, produced by the CIDR Deployment working group in February 1996. Its title called the document an appeal, and its operative language encouraged cooperation. That wording mattered. The document did not say that publication itself cancelled an assignment. It asked an organisation to decide that space was unnecessary, undertake any renumbering, and request a registry change.

The printed return form compressed the visible administrative step. It named the prefix and asked for a reserved registry contact. Even a successful registry mutation would have established only a bounded fact: the registry no longer represented the former assignment in the same way. It would not show that a router had withdrawn the prefix, that an internal application had stopped using it, that a partner had removed an allowlist entry or that the returned range was ready for another network.

This is the distinction hidden by the simple verb return. An allocation record, an operational inventory, an authorised request, a registry transaction, a route withdrawal and a later reallocation belong to different systems. They can occur in order, but one does not contain the evidence of the next.

“Not advertised” was never the same as “not used”

The RFC placed two pressures beside each other. IPv4 space was finite and unevenly used. Core routers also had finite capacity for an expanding set of routes. Reclaiming an assignment could help the first problem. Aggregation and disciplined advertisement could help the second. A single action did not necessarily solve both.

The document estimated that before RFC 1466 and CIDR, about 50,000 networks had been assigned, while only 30–40 per cent of the sites were connected to the global Internet and advertising their networks. Those numbers were a contemporary diagnosis, not a modern census. More importantly, the RFC itself described registered networks operating privately and networks sitting behind application gateways or network address translators. A prefix absent from the global table could still be carrying internal traffic.

Silence therefore had several possible meanings. The assignment might truly be abandoned. It might support a private network. It might be reserved for planned growth. It might appear only through a larger aggregate. It might be filtered from the observer’s vantage point. Route data could identify a question; it could not decide organisational need by itself.

Early generosity left several kinds of surplus

RFC 1917 traced inefficient holdings to the way the network had grown. An organisation might first receive several small networks, then a Class B after unexpected expansion, leaving the earlier assignments behind. Providers sometimes distributed Class B space according to service level rather than demonstrated need. Mergers and parent-company relationships combined address plans that had been created separately. Vendor manuals sent customers toward public assignments even for machines that never needed global connectivity.

The RFC also recorded a dramatic precedent: four previously assigned Class A networks had returned to IANA in May 1995. It did not name them in that passage or explain the evidence by which their old use was cleared. The statement supports a historical fact attributed to the RFC; it does not support reconstructing four unrecorded operational stories.

For sites that would remain isolated, the appeal pointed to the private prefixes then specified by RFC 1597. RFC 1918 replaced that document and made the bargain explicit. Private addresses have no global meaning and must not leak into inter-enterprise routing, but moving between private and public addressing can impose substantial renumbering work. Returning public space replaced one scarcity problem with an internal transition that still needed ownership and evidence.

Providers could see routes, but not every reason to keep an address

The provider section proposed several forms of cooperation. A customer using a contiguous lower portion of a large assignment might let its provider use upper portions for other customers while the provider continued advertising one aggregate. Providers should release unused networks outside their normal blocks, encourage new customers to renumber from old non-aggregated space, cooperate on aggregation and verify the routes sent upstream.

These proposals matched the provider’s control surface. A provider could inspect its own configurations, see which routes it originated and change customer routing arrangements. It could not, from that view alone, attest that an engineer had authority to surrender the organisation’s assignment or that no internal system depended on the range. “Good Internet citizenship” supplied an incentive; it did not merge the provider, assignee and registry into one decision-maker.

Later documents separated voluntary stewardship from registry power

RFC 2050 later used stronger language. It described assignments as valid while their criteria continued to be met and reserved an IANA right to invalidate assignments when the requirement no longer existed, with reasonable registry efforts to notify the organisation. That framework should not be projected backwards onto RFC 1917. An appeal for voluntary return and a policy for invalidation are different institutional instruments even when both seek better utilisation.

RFC 7020 made another boundary explicit. Allocation-pool management, hierarchical allocation and registration accuracy are goals of the Internet Numbers Registry System. Whether an address is announced, and how it is announced, are operational matters outside that system’s scope. Registration can protect uniqueness and record delegation without proving routing or service state.

The current procedure for a legacy /8 illustrates the layered path. IANA directs a record update to the relevant RIR; the RIR applies its data-maintenance procedure and notifies IANA if the top-level registry must change. A complete /8 return passes through the RIR before reaching IANA’s unallocated pool. That is a procedure published years after RFC 1917, not a description of the 1996 email form.

The modern recovered-address registry is equally bounded. It records ranges returned by RIRs, recovery dates and allocations made under the global post-exhaustion mechanism ratified in 2012. It does not identify every original end user, why each range came back or whether RFC 1917 influenced the decision. A recovered-pool row is strong evidence of registry state. It is not a complete biography of the address.

Sources