Summary

  • Draft 1, submitted on 27 February 2019 and announced on 7 March, proposed replacing AFRINIC’s cumulative ASN eligibility conditions with two alternative routes: multihoming, or a need to interconnect with another autonomous system. It remained Under Discussion and did not become an operative policy.
  • The proposal correctly separated the function of a globally unique routing identity from the purchase of a second physical upstream. That distinction could matter to a remote or smaller operator using provider-independent space, joining an IXP, arranging a private interconnection or preparing for a future provider change.
  • The reform was incomplete. “Need to interconnect” did not itself define when public uniqueness was necessary, while Draft 1 retained an RFC 1930 reference whose section 5.1 said a single-homed site did not need a unique ASN. AFRINIC’s 8 April staff assessment made that conflict explicit.
  • The defensible boundary is neither compulsory multihoming nor automatic issuance. AFRINIC may verify a concrete need for a globally coordinated public ASN, keep accurate records and direct local-only uses to private numbers. As a private registry, it may not dictate redundancy or treat its process as regulatory authority over network architecture.

The applicant’s problem: paying for topology to obtain identity

Consider an operator serving a town where one carrier is the only practical source of transit. The operator may run its own address space, expect to exchange traffic at an Internet exchange point, or maintain a private interconnection with another network. It may need its routes to carry a stable public identity that is not merely an internal label chosen by its current transit provider. What it may not be able to do is buy a second physical upstream. The second circuit may not exist. It may require a long access build, new equipment and recurring transit payments.

Even where it exists, it may be economically irrational for the operator’s size and traffic.

Under the inherited AFRINIC wording reproduced on the archived Draft 1 page, this distinction was obscured by the logical structure of the eligibility test. An applicant had to meet all four listed conditions. The conditions included having a unique routing policy and being a multihomed site, or having a near-term plan to become one, alongside good-standing membership. The conjunction mattered. A network could explain why its public routing identity needed to remain distinct and still fail because it had not purchased the topology that the rule used as proof.

Draft 1’s core redline was small enough to sound administrative. A member in good standing would be eligible if it was multihomed or could show a need to interconnect with another autonomous system; a near-term plan to meet either route would also count. The proposal also deleted an existing introductory subsection and consolidated and renumbered Section 7. Those editorial rearrangements were part of the text, but the consequential move was from cumulative conditions to alternatives. One word in the logical structure—effectively changing an AND into an OR—separated physical redundancy from another possible basis for public identity.

The documentary sequence must be kept precise. AFRINIC’s archived proposal page says Version 1.0 was submitted on 27 February 2019, names Jordi Palet Martinez as author and labels the proposal Under Discussion. AFRINIC Communication announced receipt on 7 March, summarized its aim as removing multihoming as a prerequisite for ASN assignment and invited comment on the RPD list. Submission and public announcement were separate acts on separate dates. Neither was adoption.

The identifier itself also survives in two forms. The archived page visibly prints AFPUB-2019-ASN-DRAFT01. The staff assessment circulated in April prints AFPUB-2019-ASN-001-draft-01, while the later proposal chain and the normalized identifier for this analysis use AFPUB-2019-ASN-001-DRAFT01. The discrepancy has no demonstrated substantive effect, but erasing it would imply a documentary neatness the record does not possess. It is better treated as a recordkeeping inconsistency than explained by speculation.

What problem did the alternative route try to solve? The strongest contemporaneous answer came through examples, not abstraction. Fernando Frediani argued that a network might have stable provider-independent address space even while using one upstream. It might want the freedom to change providers without changing its routing identity. It might connect to an Internet exchange point, where its relationship to multiple peers is not captured by counting paid transit carriers. It might maintain a private interconnection with another autonomous system. It might also be preparing for a later upstream choice.

In each case, “single transit upstream” and “no need for a globally unique public routing identity” were not necessarily the same proposition.

That qualification is essential. It does not follow that every single-homed site needs an ASN. Jaco Kroon raised the cleanest objection in the March discussion: if a site has one upstream, that upstream can originate or carry the route, while the site uses a static route and does not need its own autonomous-system identity. This was not a trivial defence of the status quo. It identified the registry’s legitimate concern. Public ASNs are coordinated identifiers, not badges of maturity. If an operator’s routing is wholly contained within one provider relationship, a static arrangement or a private ASN may be adequate.

Draft 1 retained that side of the policy. It did not abolish eligibility review. It preserved guidance that sites not requiring a unique ASN should use private autonomous-system numbers. It also retained evaluation under RFC 1930 or its successors. Those choices show that the proposal was not an invitation to issue a public number to anyone who preferred one. Its intended distinction was between physical multihoming as a compulsory prerequisite and other circumstances that could produce a real interconnection need.

Private ASNs are important to that boundary. Reserved private-use ranges let networks coordinate identifiers inside a limited administrative environment without consuming a number whose uniqueness must be recognised across the public routing system. RFC 6996 documents those reserved ranges. Where uniqueness is only local, the private tool fits. The question becomes different when the identity has to remain unambiguous across independently operated networks and public routing relationships. The registry’s task is to tell those cases apart, not to assume that every public need comes in the physical form of two transit contracts.

The economic error in the inherited proxy was therefore not simply that a second upstream costs money. Many useful network choices cost money. The error was tying that expenditure to a different service: the coordination of a globally unique identifier. The cost of multihoming can include recurring transit charges, access circuits, ports, routers, configuration, monitoring and operational skill. In some places, geography makes the second route scarce or unavailable. None of those facts proves that the applicant’s public ASN request is sound.

But none is a proper reason to require redundancy when the applicant can demonstrate the distinct coordination need directly.

For a large operator, an extra transit relationship may be ordinary. For a small network, the same formal condition can consume a much larger share of revenue and staff time. Heng Lu’s poverty-penalty analysis is useful here because it directs attention to fixed compliance costs. A rule can look equal—every applicant must show two connections—while its burden is radically unequal. The incumbent with engineers, bargaining power and metropolitan fibre treats the condition as paperwork. The rural network or new entrant must reshape its business, wait for construction or buy a service unrelated to the immediate reason it needs a number.

That burden can also deepen supplier dependence. Stable provider-independent addressing and an independent public ASN can make a future provider change less disruptive, where the technical routing arrangements genuinely require them. If the operator must first obtain a second provider in order to secure that identity, the rule can lock the operator more tightly to the provider whose dependence it is trying to reduce. The test becomes circular: independence is available only after the applicant has already purchased the market conditions associated with independence.

The sealed public record does not establish that a named applicant was denied or deterred under this rule. It does not count how many networks faced the problem, calculate the cost of a second upstream or measure benefits after later implementation. Those absences matter. The mechanism is credible and the examples are technically intelligible, but a credible mechanism is not a measured outcome. The careful claim is that the inherited condition could impose topology cost unrelated to identifier need, and that the March debate identified concrete classes of operator for whom the mismatch could arise.

The word “could” should not weaken the analysis; it locates it. Eligibility rules act before many losses become visible. A small operator may not file a doomed application. A planned exchange connection may be deferred. A provider-dependent arrangement may persist because the prerequisite makes an alternative too expensive. Without applicant files, checklists and comprehensive request data, it would be irresponsible to convert those possibilities into a casualty count. It would be equally irresponsible to ignore the design of a rule merely because deterred applicants are hard to enumerate.

The March discussion sharpened the technical object of the test. Owen DeLong proposed a simpler formulation centred on a unique routing policy or an interconnection using an exterior gateway protocol that required a unique ASN. Sander Steffann supported that direction. Their intervention matters as evidence of the problem participants were trying to solve: “interconnection” by itself might describe relationships that did not require globally unique identity, whereas an exterior routing relationship or a genuinely distinct policy made the coordination need more legible.

That proposed language in the discussion was not Draft 1’s published text. It cannot be silently substituted into the redline. Its significance is diagnostic. Once physical multihoming ceased to do all the filtering work, contributors immediately searched for a criterion closer to the function of the identifier. Their exchange reveals why the first alternative was both directionally correct and operationally incomplete.

The difference between an IXP and a second transit supplier illustrates the point. At an exchange, a network may establish routing relationships with several independent participants while buying transit from only one upstream. Counting transit carriers would label the network single-homed, yet its exchange participation could require a stable identity visible to multiple autonomous systems. Conversely, a physical connection to one other network does not automatically prove that a public ASN is necessary; the relationship might be served by a private number within a controlled arrangement.

“Number of cables,” “number of upstream contracts” and “scope over which identity must be unique” are related facts, not interchangeable tests.

A private interconnection presents the same distinction in smaller form. Two independently administered networks may exchange selected routes under policies that cannot be represented solely through the applicant’s transit provider. That can support a public-identity need even if the applicant buys Internet transit from one carrier. Yet the registry still needs evidence: who is interconnecting, what routing relationship is intended, why a local private identifier would not suffice, and when the arrangement is expected to operate. Draft 1 opened this path without fully defining the evidentiary threshold.

Provider-independent space adds another layer. An address block can remain stable across providers, but its global reachability and policy may be represented in different ways. Frediani’s argument was that stable space, future upstream choice and interconnection needs could make an ASN useful before a second transit link was active. Kroon’s answer reminds us that “useful” and “necessary for global coordination” are not identical. A well-designed test would require the applicant to connect the desired stability to an actual public routing arrangement, not merely assert that independence is generally desirable.

This is why the benign defence of the inherited rule must be stated at full strength. Multihoming was a conservative engineering heuristic. It often signalled that a network operated an autonomous routing policy visible to more than one external system. It gave staff a relatively objective fact to verify, discouraged unnecessary public numbers and aligned with familiar RFC 1930 guidance. A static route really is enough for many single-provider customers. Replacing a clear heuristic with an undefined claim of “need” could enlarge inconsistency, applicant effort and staff discretion.

The answer is not to deny those benefits. It is to separate evidence from entitlement. Physical multihoming can remain strong evidence that a unique ASN is needed. It should not be the exclusive admission ticket. The registry can accept other evidence that addresses the same coordination question: an IXP arrangement, a documented private interconnection across independent networks, a distinct exterior routing policy, an operational requirement tied to provider-independent announcements, or another concrete situation in which local numbering cannot keep the identity unambiguous.

The operator’s case should also be assessed proportionately. A network planning an interconnection should not have to build the entire relationship before it can obtain the number needed to configure it. Draft 1 recognised near-term plans under either eligibility route. But a plan must be more than an indefinite ambition. Dates, counterparties, intended routing relationships and an explanation of why a globally unique identifier is needed can make a prospective case reviewable without letting the registry demand capital expenditure as proof of sincerity.

That proportionality is especially important where lead times are long. A remote operator might coordinate an exchange port, transport service, equipment installation and address announcements in parallel. Requiring completed multihoming before assignment can create a sequencing trap: the operator needs the ASN to complete routing configuration but needs the completed topology to obtain the ASN. Evidence of a real, near-term interconnection resolves the sequence more directly than buying an unrelated second transit service.

The practical case for Draft 1 was therefore not that redundancy lacks value. Redundancy can improve resilience, diversify suppliers and reduce some failure risks. Operators may sensibly choose it. The point is institutional: the registry’s identifier service should not decide whether the operator must buy it. Once the registry turns a resilience option into an eligibility condition unrelated to the actual uniqueness need, a technical recordkeeping rule begins to govern investment and architecture.

NRS’s description of AFRINIC as a membership-based number registry helps keep the institution in scale. AFRINIC coordinates Internet number resources; it is not a planning ministry for African networks. LARUS’s operator-continuity analysis adds the corresponding operational insight: delays and eligibility interpretations at a registry can influence real infrastructure decisions even when the registry is not the operator. BTW’s analysis of a thinner coordination architecture frames the alternative: preserve a reliable visibility and uniqueness layer while reducing discretionary control over how networks are built.

Those sources do not prove particular 2019 applicant outcomes, and they should not be used as substitutes for the archived proposal or mailing-list record. Their role is analytical. They explain why a requirement located in a registry form can have consequences beyond the form, and why those consequences should remain connected to the registry’s narrow function.

At the operator level, then, the decisive question is simple to state but demanding to administer: does this network need an identity that must be unique across the public coordination domain? Multihoming may answer yes. An IXP, private exterior relationship or stable independent announcement may also answer yes. A preference for prestige, a local-only topology or a customer connection adequately served by a static route may answer no. Draft 1’s achievement was to stop treating the first fact as the only imaginable proof. Its defect was not yet saying clearly enough how the other answers would be judged.