Summary

  • RFC 2008 treated address ownership and address lending as routing-policy models, not as legal conclusions: keeping a prefix and obtaining worldwide acceptance for its route were separate outcomes.
  • Lending tied an address to the service relationship that supplied aggregation. Leaving that relationship brought a grace period, renumbering and return, while keeping the old prefix could create a more-specific route that other networks were free to reject.
  • The document did not eliminate cost. It allocated cost among renumbering, global routing state, partial reachability and mediating gateways.

The difficult moment came after both obvious parties had said yes. The old provider was willing to let a departing customer keep part of its block. The new provider was willing to announce that more-specific prefix. From the customer’s perspective, portability appeared settled.

RFC 2008 introduced a third party: every other provider whose routers would have to carry the exception. One could filter the longer prefix. Another could be unable to accept it. The customer would then reach some networks but not others. A bilateral accommodation had not become a worldwide service.

That distinction is the centre of Yakov Rekhter and Tony Li’s October 1996 BCP. It is not another lesson in CIDR notation. It is an account of which promise came from which actor.

The value came from the service

RFC 2008 began with an austere description. An IPv4 unicast address was a finite integer used by network-layer routing. In the Public Internet, its notable value was the ability to interact with the routing service and exchange data with the rest of the network. The document therefore put the causal arrow from the operating Internet to the useful address: if the routing system stopped, the allocation policy would not preserve the address’s Internet function.

That was not a claim that addresses could never acquire contractual, commercial or identity value. It was a boundary on the RFC’s technical object. An entry could preserve uniqueness. It could identify an assignee. It could not command every router to accept a route.

RFC 1518 had already explained the scaling bargain. A provider could allocate longer prefixes from its own block and advertise one shorter aggregate to other providers. The direct customer and provider might see little difference between an independent prefix and a provider-derived one; much of the saving appeared elsewhere, in the tables and exchanges of networks that no longer had to retain a separate route for that customer.

The benefit was distributed. So was the power to withhold it.

Ownership was a policy label

RFC 2008 used “address ownership” in quotation marks for a defined allocation policy. Once allocated, the space remained with an organisation as long as it wanted it and would not be allocated to another. Such addresses were often called portable. The operational expectation attached to that label was stronger: the organisation expected to use the same numbers wherever it connected.

The RFC did not convert that expectation into a general property doctrine. It asked whether the routing system of 1996 could carry the consequence. If every site moved freely while keeping a separately visible prefix, the topology could change without the addressing hierarchy changing with it. Each move could punch another exception through an aggregate.

For organisations able to cover a sufficiently large fraction of Internet destinations and express that reachability through one prefix, the authors regarded ownership as reasonable. Such a prefix could plausibly remain throughout the default-free routing system. For ordinary sites that did not aggregate enough destinations, the same treatment at universal scale was, in their judgment, technically infeasible.

The distinction was not moral worth. It was compression. How many destinations could one routing entry truthfully cover?

Lending attached the number to the aggregator

The alternative was explicit “address lending.” During the loan, the lender could not lend the same addresses to anyone else. The conditions were supposed to say that the allocation returned when borrower and lender were no longer contractually bound and the lender could no longer aggregate it.

Where a provider-associated registry made the loan, RFC 2008 expected the provider to arrange enough aggregation for Internet-wide connectivity. The service agreement and the address loan were coupled. If the subscriber moved, the loan ended; the subscriber obtained new addresses and renumbered into the new provider’s block.

The document made the exit interval concrete. It strongly recommended a grace period after disconnection: at least 30 days to reduce disruption, but no more than six months to contain routing overhead. Those bounds reveal the compromise. Immediate return ignored operational change. Indefinite exception transferred the old customer’s continuity cost into other networks’ routing tables.

Nor was lending necessarily one layer deep. A borrower could lend portions onward. Ending one upstream loan might therefore disturb several downstream organisations, an effect RFC 2008 said was difficult to know in advance. The apparently simple contract had a dependency tree.

No policy included global reachability

The document’s sharpest warning concerned providers that promised no renumbering without an explicit lending policy. Even if both the former and new provider agreed to preserve and propagate the prefix, other networks were not parties to that promise. Internet-wide connectivity might still require renumbering, or require affected correspondents to buy connectivity from providers willing to carry the exception.

This was a precise limit on registry evidence. Allocation did not guarantee announcement. Announcement by one provider did not guarantee propagation. Propagation did not guarantee acceptance. A route visible from one vantage point did not prove universal reachability.

RFC 2050, published the following month, carried the loan logic into registry guidance. It recommended returning provider-issued addresses on termination and allowing sufficient time before reuse. It also warned that addresses issued directly by registries were the least likely to be routable across the Internet. That later text corroborates the period’s operating concern; it is not evidence that every provider adopted the rule or that the rule remains current.

RFC 7020 later described the separation in institutional terms. Hierarchical allocation could improve the likelihood of scalable routing, but actual announcement and advertisement were operational matters outside the Internet Numbers Registry System. A register and a route remained different instruments.

Choosing which cost to carry

RFC 2008 offered no costless branch. Portable, non-aggregatable space consumed routing state, and the organisation injecting it might eventually be charged the full cost. Lending preserved aggregation but made renumbering part of exit. Limited connectivity could keep a route local. Mediating gateways such as application gateways or address translators could avoid some global exceptions and some renumbering, but introduced awkward constraints of their own.

Renumbering was already known to be a serious price. RFC 1900 called it costly, tedious and error-prone, with limited tooling. RFC 2008 added firewall coordination and warned against using unauthenticated addresses as identity. A new prefix was not merely a new row; security rules inside and outside the site might have to move with it.

Lu Heng’s later writing makes the historical tension visible without rewriting it. The Bill of Rights of Uniqueness Coordination argues that a registry record describes reality rather than creates it, while demanding portability and operational continuity. On LARUS One asks what it costs when customers, banks, partners and security systems remember a number. On the Nature of IP Addresses distinguishes the original identifier from the asset-like value produced by markets and dependence.

Those are contemporary positions, not evidence of what RFC 2008 legally decided. Their value here is diagnostic. The RFC correctly showed that registration could not manufacture worldwide routing. Later economic dependence shows why telling a network simply to renumber can allocate a much thicker cost than the address policy records.

The durable lesson is narrower than either slogan. An address record is not a route. A route from one provider is not global acceptance. A portable expectation is not a universal obligation. A loan’s end is not completed renumbering. In 1996, the Internet’s addressing debate was already a debate over who would carry the exception—and who would pay when the number, the topology and the business refused to move together.

Sources