Summary

  • Izumi Okutani’s August 2004 APNIC proposal showed that the NIR IPv6 per-address fee could count an existing allocation again when the registry expanded it. In her illustration, a staged path to a /29 accumulated 78,809 fee units; requesting the /29 at once counted 37,641.
  • The APNIC 18 NIR SIG accepted her narrower remedy: charge the difference between the old and new allocation, not the entire new prefix. Its minutes record support without objection. The Executive Council endorsed the proposal in November 2004; APNIC announced the changed fee-schedule text in March 2005.
  • A second proposal at the same meeting sought a broader cap on NIR fees and did not reach consensus. Okutani’s success repaired the cost of growing. The argument over the overall fee structure continued.

A fee for the space you already had

A ledger should be able to remember what it charged yesterday. The IPv6 formula Izumi Okutani put before APNIC in 2004 had a more adventurous memory. When a National Internet Registry, or NIR, received an expanded allocation that contained its earlier block, the per-address fee calculation used the size of the new prefix, regardless of the previous allocation. Okutani quoted that rule from APNIC’s then fee schedule in her proposal. In accounting terms, the new envelope swallowed the old one and the bill treated the contents as fresh.

This concerned a particular layer of APNIC’s system. APNIC allocated space to NIRs; NIRs served networks in their own economies. Okutani was working at JPNIC and presented the problem from that seat. Her paper says the NIR could absorb the redundant charge or pass a similar method to its members. JPNIC had applied the same scheme and, she wrote, could not give a reasonable justification for it. That sentence has more life in it than most policy papers. It describes the awkward moment when an intermediary must explain a charge it cannot defend to the customer who actually wants to build the network.

There is a difference between charging for more address space and charging again for space already included in a prior allocation. The first can be a policy choice. The second is a defect in the way the policy counts. Okutani’s proposal attacked the count. It asked the invoice to know subtraction.

The invoice had an address history. It chose to forget it.

The old sentence in APNIC-081 did more work than its modest size suggested. An IPv6 allocation was priced using the number of /48 equivalents associated with its prefix and an HD-ratio of 0.80. That is a utilization model: the invoice did not simply multiply every possible address in the prefix by a flat price. The model produced a fee-calculation figure for the prefix. Yet when the prefix grew around a block already allocated, the rule instructed the calculator to use the size of the new total prefix, regardless of the earlier one. The problem was buried in that final instruction. APNIC could know exactly which range an NIR already held and still send the new calculation through as though the earlier range were arriving for the first time.

This is easier to see with paper than with hexadecimal. Draw a /32 on one sheet. Put the sheet inside a larger envelope marked /31. The old /32 is still there; nobody has created it again. The operator has gained the other half of the /31. If the fee schedule treats the entire /31 as a new charge, it collects against the sheet already paid for. Move that envelope into a /30, and the same sheet appears in a third charge. Okutani’s proposal diagram makes precisely this point with nested boxes. It is a diagram of memory failure, drawn by someone who had to help administer the rule.

APNIC’s 2005 draft fee schedule is useful because it places the repaired IPv6 clause next to the IPv4 clause. The IPv4 section already said that an allocation containing a previous block includes only the additionally allocated addresses in the new fee calculation. IPv6 used a different utilization measure, but the underlying question was the same: which part is newly arriving? Okutani did not ask the community to abolish the IPv6 formula. She asked it to apply the formula to the change between two known states. Once the schedule used the difference in /48-equivalent utilization figures, an expanding prefix no longer made yesterday’s allocation look new again.

The wording also reveals how an apparently neutral table can distribute advantages. A customer who arrives asking for a /29 can be charged on the /29 figure once. A customer who reaches /29 through several supported expansions can be charged on each intermediate total unless the table keeps track of what was counted before. Both end with the same nominal allocation. Their different histories, not their final holdings, drive the old bill. That is why the counterexample was so effective: it held the destination constant and changed only the path.

A policy defender could still argue about the proper rate, about cost recovery, or about which utilization ratio to use. None of those arguments explained why the same destination should generate two totals under the same published table.

The people between APNIC and the network mattered. The schedule described per-address fees for confederation members, including NIRs. An NIR received allocations from APNIC and served its own members. Okutani wrote that JPNIC applied the same method and could not give a reasonable justification for the redundant charge. That is unusually candid administrative language. The person facing an operator does not get to answer, “the formula says so,” indefinitely. When the upstream invoice counts old space twice, the intermediary either absorbs a cost or passes a similar calculation on.

Both choices leave the intermediary explaining a rule it did not invent. Okutani’s case began at that uncomfortable point of contact, where a spreadsheet cell becomes somebody’s conversation with a member.

The wrong prize for being careful

The original prop-022 text makes the perverse incentive hard to miss. Imagine a network beginning with a /32, then expanding to a /31, a /30 and eventually a /29. Under the old method, the fee calculation counted 7,132 units at the first step, then 12,417, then 21,619, then 37,641. Add the staged charges and the illustrative total is 78,809 fee units. A network that asked for the final /29 in a single first request faced 37,641 fee units in the same illustration.

The proposal expresses these as illustrative fee-calculation units. The difference is 41,168 units. It exists because the earlier space returns inside each new prefix and gets counted again. The calculation isolates a defect in the rule itself; APNIC’s published record supplies no network-level invoice series.

The direction of the incentive matters. A registry concerned with responsible distribution might expect operators to begin with what they need and expand when demand arrives. This fee rule did the opposite in the example: it made the gradual route cost more than the large initial request. An operator could choose a large block early to avoid being repeatedly charged for prior space. The published case measures the incentive in the rule itself: under its example, caution carried an extra charge.

Okutani’s replacement calculation kept the size-based method but moved the charge to the increment. A /32 would still be counted when first allocated. On expansion to a /31, the relevant units would be 12,417 minus the 7,132 already counted. At the next step, the calculation would subtract the /31 amount from the /30 amount, and so on. The staged path would then add up to 37,641 units, exactly the figure for requesting the /29 once. The arithmetic became indifferent to the route taken to the same final allocation.

Her paper also gave a more concrete rate example. For a /30 that included an existing /32, it used the difference between 21,619 and 7,132 /48-equivalent units at an illustrative rate of three US cents, producing a $434.61 per-address fee. The worked calculation used the document’s “Very Large” member tier and its 2004 assumptions. Its point was the incremental formula.

A four-line sum with an institutional punchline

The four-stage path deserves to be read slowly. At /32, the illustration counted 7,132 units. The /31 figure was 12,417. The /30 figure was 21,619. The /29 figure was 37,641. Under the old instruction, the ledger wrote down four full figures and added them: 7,132 + 12,417 + 21,619 + 37,641 = 78,809. Under Okutani’s replacement, it wrote 7,132 first, then the change to 12,417, then the change to 21,619, then the change to 37,641. The intermediate entries cancel when added. The result is 37,641. No operator needs a new routing design to make that happen. The fee schedule simply needs to recognize the old allocation in its calculation.

The second path, a /29 requested from the beginning, entered the ledger once at 37,641. Under the old rule, that was the better financial path in the illustration. The gap between 78,809 and 37,641 is 41,168 fee units. The published proposal table is an example of how the rule behaved, with the unit multiplied by the applicable per-address rate. The point is the shape of the choice it presented. A cautious operator could be told to start with a block that fits demand, and then find that the pricing system favored asking for the eventual larger block immediately. It is possible to write a conservation policy with one hand and a fee incentive against gradual growth with the other. The operator encounters both at the same counter.

The “Very Large” example shows that Okutani also knew the fix had to survive an actual table entry. The March 2005 draft listed the corresponding /48-equivalent utilization figures for /32 through /29, and it listed rates by membership tier. It showed a /30 at 21,619 units multiplied by $0.03, or $648.57, when requested without the specified earlier /32. With that /32 already inside it, the worked new-clause example subtracts 7,132 and arrives at $434.61. These are two different allocation histories under the document’s example rate.

The dollar figures are demonstrations of the clause; the numbers do not announce what every NIR or every network paid. The valuable part of the demonstration is that a clerk could reproduce it from the table without asking what a policy author meant by “fair.”

The unit choice is worth lingering over because IPv6 sizes can make ordinary intuition desert the room. The schedule’s /32 line listed 65,536 possible /48s and a utilization figure of 7,132 under the HD-ratio assumption. Its /29 line listed 524,288 possible /48s and a utilization figure of 37,641. Neither 7,132 nor 37,641 is a count of invoices or customers. They are numbers the fee formula derived from the size of an allocation. Okutani used those same numbers, rather than substituting a more emotionally satisfying example, because the defect lived inside the published arithmetic. The repair had to speak the language of that arithmetic.

Readers need not memorize an HD-ratio equation to follow the story; they need to notice which of the table’s figures the invoice counts a second time.

For someone filing an expansion request, that difference would not be abstract. A request made today can change the cost of a request made years later if the schedule adds whole-prefix charges. A network planner may have sound engineering reasons to start smaller. The old example made that planner face a financial disadvantage for choosing the staged path. Okutani did not claim to have measured how many operators changed their requests because of the rule. She showed that the rule itself priced the two paths differently. A government does not need to count every driver who avoided a toll road to prove that the toll exists.

Here, the toll was printed in APNIC’s fee schedule.

The fee schedule carried other charges and categories: annual membership fees, membership tiers, an initial resource application fee and per-address charges for confederation members. This context matters because prop-022 did not flatten all those lines into one bill. The 2005 draft’s membership tier table placed a member’s tier against address holdings. Its IPv4 clause used actual additionally allocated addresses for an expansion. The IPv6 clause used a utilization number derived from the HD-ratio. Okutani’s surgical change touched the expansion sentence in section 3.4.3.

A narrow sentence can be powerful because the accounting system around it stays recognizable. The Secretariat did not need to devise a new institution to stop repeating a charge.

There is a satisfying smallness to the remedy. Grand policy disputes tend to produce grand words: sustainability, fairness, regional development. This one produced a minus sign. The minus sign did not settle whether NIR fees were too high, whether a particular membership tier was appropriate or how APNIC should balance revenue with services. It answered a question that could be decided from the schedule’s own logic: if the invoice already counted one range, how much new range is inside the expansion? Okutani gave the answer in units that the existing table could understand.

The meeting accepted the narrow repair

At APNIC 18 in Fiji on 2 September 2004, Okutani presented the case to the NIR SIG. The meeting minutes summarize the issue plainly: a conservative ISP that starts small could be financially disadvantaged by the charge on later expansion. The Secretariat had examined the likely effect on APNIC’s operating expenses and said it would not be significant based on the upgrades that had happened so far. Participants called the anomaly unfair and reasonable to address. The chair asked for a show of hands, saw support without objections, and declared consensus.

The show of hands began a longer adoption sequence. The action item remained conditional on the rest of APNIC’s policy process. A final call for comments then described the agreement as removing multiple fees for the same IPv6 address range allocated to NIRs. It set 12 November 2004 as the comment deadline. APNIC’s proposal register records Executive Council endorsement on 19 November. Its implementation date is 14 March 2005.

The 14 March APNIC notice says section 3.4.3 of the draft member fee schedule had been amended to remove multiple charging when existing allocations were expanded. It also sought editorial comments on whether the draft accurately reflected the already agreed policy. The public record therefore has several distinct acts: Okutani proposed the correction; the NIR SIG found consensus; the EC endorsed it; the Secretariat changed the schedule text and exposed that text for final review. Each step moved the billing rule closer to the invoices NIRs would receive.

The room could count the votes. The fee table still needed a line edit.

The NIR SIG record starts with a small biographical fact that belongs in this story. At the same meeting, the SIG selected two co-chairs, David Chen of TWNIC and Okutani of JPNIC. The minutes say both described hostmaster and training experience and an interest in contributing to the community. Then, later in the agenda, Okutani presented her fee correction as a JPNIC proposer. The public record gives her both a working role in the community and ownership of this particular text. It does not require a myth of a solitary hero discovering arithmetic that everyone else had missed. The proposal succeeded because she brought a precise case to a room with a process for testing it.

The process took more than a show of hands. The Secretariat had considered whether stopping the redundant charge would matter to its operating expenses. The minutes describe the likely effect as not significant based on the upgrades seen so far. That assessment helped distinguish the correction from a wholesale change in revenue. Participants recognized the anomaly as unfair; the chair called for support and saw no objections. Yet the meeting’s action item was explicitly conditional on the remaining stages of the policy process. Consensus in a session is a decision to advance, not a magic eraser applied instantly to every invoice.

APNIC’s records make the distinction legible by preserving the later comment call, Executive Council endorsement and implementation date.

The September final call translated the meeting’s argument into a community question. It described the measure as removing multiple charges for the same IPv6 range allocated to NIRs and invited comments until 12 November. The proposal register then records the Executive Council’s endorsement on 19 November. Those dates show the path from an author’s counterexample to an adopted rule. They also show why an archive can tell this story more reliably than a retrospective quote. Every stage has its own verb and its own institution: propose, discuss, call for comments, endorse, implement.

Implementation had its own calendar. The proposal register gives 14 March 2005. APNIC’s announcement on that date introduced revised draft fee-schedule wording and asked for editorial comments through 11 April. The draft document identifies itself as version 008 of APNIC-081, dated 14 March, with draft or provisionally active status. Section 3.4.3 contains the operative replacement: use the difference in the /48-equivalent utilization figures between the previous allocation and the new one. Reading all three records avoids compressing proposal, approval and document publication into one triumphant afternoon in Fiji. A correction can win a room in September and still require months of orderly institutional work before it appears in the fee text.

The distinction is also part of Okutani’s achievement. She did not merely point to a bad outcome. She supplied replacement language that could be carried across those stages with the underlying calculation intact. The implementation notice could then ask a narrow editorial question: did the draft faithfully express the policy already agreed? That is the point at which a good argument has become an administrable instruction. A member checking a later bill needs the instruction, not a memory of applause.

The larger fee fight stayed larger

The same NIR SIG minutes record a different proposal from Toshiyuki Hosaka of JPNIC to change the wider NIR fee structure by setting caps. It raised questions about APNIC revenue, fairness between direct and indirect members, and whether a cap would move costs to other participants. The debate had no consensus on that proposed cap; the chair instead found support for a working group to examine NIR fees. The distinction matters. Okutani’s correction could be accepted because it stopped charging twice for the same range without requiring the room to agree on the total revenue model.

The other proposal tested the fee level and exposed disagreement. Prop-022 was narrower and cleaner: whether two allocation histories ending at the same block should yield the same total per-address fee. The room could say yes to that equality while continuing to argue over how high the fee should be in the first place.

The larger discussion was hardly a footnote. According to the same meeting minutes, Hosaka proposed caps at a /14 for IPv4 and a /28 for IPv6. People in the room argued about APNIC’s stable service revenue, the share paid by direct members and members who came through an NIR, and the possibility that a cap would favor especially large requests. The Secretariat’s initial analysis pointed to a substantial revenue reduction under the proposed cap. Participants also raised the prospect that a large ISP could join APNIC directly if the NIR path became too expensive. One comment observed that the NIR charging structure was hard for ordinary members to understand. Another wanted the whole structure reconsidered. The argument touched price, membership design, service funding and incentives all at once. An elegant subtraction cannot settle a debate with that many owners.

The chair found support for studying alternatives and called for a working group, but did not find consensus on Hosaka’s cap. That outcome sharpened the meaning of Okutani’s win. She had identified a duplicated charge for an already allocated range. The room could remove that duplication without settling who should bear APNIC’s total costs. The two proposals were presented side by side, by people from the same national registry, and the minutes recorded different results.

Any account that calls the meeting either a sweeping fee reform or a failed fee reform loses the more interesting fact: the community could repair a demonstrable calculation error while leaving the revenue bargain open.

The minutes even show why that bargain was harder. A fee cap changes what APNIC collects from some large allocations and invites the next question: whose fee or service changes to fill the hole? A repeated charge on an existing range is different. Its defender must explain why the range counts as new again. For prop-022, the Secretariat could also say the observed upgrade pattern suggested limited operating-expense impact. That made the narrow remedy easier to accept, though it was the arithmetic, not the small estimated budget effect, that supplied its logic. Okutani chose a claim the room could verify on the page.

This is a useful portrait of her as a policy author. The available records show a JPNIC staff member using a public fee schedule, a diagram of nested allocations and a worked numerical example to correct an upstream rule. They do not need invented backstage scenes or claims about her private motives. The proposer’s discipline is visible in the document: quote the clause, display the bad result, write the replacement clause, display the corrected result, state who is affected. For a policy forum, that is an unusually reader-friendly plot.

The villain is one sentence in a table; the resolution is a different sentence that lets the table remember.

That is also why the story belongs to Okutani rather than to an abstract spreadsheet. Her JPNIC position connected a formula at APNIC to the people obliged to explain it locally. The paper used the fee schedule’s own language and an arithmetic counterexample to show the mismatch. It then proposed a sentence the Secretariat could put into the rule: calculate the difference between the old and new allocation. The question was small enough to act on and exact enough to check later against the published text.

The cautious operator needed a mathematician to expose the duplicate charge. Okutani did the arithmetic in public. APNIC changed the line in the schedule. The ledger finally remembered yesterday.