Summary

  • RFC 10028, dated August 2026, assigns six ranges of the dynamic IPv6 multicast group ID space in its Section 3, and states in Section 6 that IANA has created a registry for them under a "Standards Action" registration policy.
  • IANA's "Dynamic Multicast Group IDs" registry prints all six ranges with the same descriptions and the same references, and prints its registration procedures as three separate lines that cover three of the six rows.

Two documents, two publishers, one act of allocation. RFC 10028: Updates to Dynamic IPv6 Multicast Address Group IDs is an IETF Standards Track document that updates RFC 3307; its header records a publication month of August 2026, and the RFC Editor's own information page labels it a Proposed Standard. The registry it names is a table on IANA's IPv6 multicast pages. Put the two side by side and the first thing that happens is a pleasant surprise. The second is a small, specific disagreement about a field labelled procedures.

What the standard prints

RFC 10028 names three authors with affiliations: N. Karstens (Garmin), D. Farinacci (lispers.net) and M. McBride (Futurewei). Section 3 carries two tables. Table 1, "Existing Allocations", is drawn from RFC 3307 Section 4.3 and RFC 4291 Section 2.7.1 and describes the space as it stood before the update, in two rows across three questions: 0x80000000-0xFEFFFFFF answers No for solicited-node, Yes for server allocation (MADCAP) and Yes for host allocation; 0xFF000000-0xFFFFFFFF answers Yes to all three.

Table 2, "Updated Allocations", is the six-row version, and it is the one the registry had to absorb:

What the standard prints
Range Description Reference
0x80000000-0x8FFFFFFF MADCAP "Defined in [RFC2730], range assigned in RFC 10028"
0x90000000-0xEFFFFFFF Unassigned (blank)
0xF0000000-0xFCFFFFFF Host allocation of SSM group addresses RFC 10028
0xFD000000-0xFDFFFFFF Reserved for Private Use RFC 10028
0xFE000000-0xFEFFFFFF Reserved for Experimental Use RFC 10028
0xFF000000-0xFFFFFFFF Solicited-Node multicast addresses "[RFC4291], Section 2.7.1"

Then Section 6, IANA Considerations, does the administrative half in two sentences: "IANA has created a new registry named 'Dynamic Multicast Group IDs' in the 'IPv6 Multicast Address Space' registry group. The 'Standards Action' registration policy is required to update the registry." One registry. One policy. Six rows.

What the registry prints

IANA's IPv6 Multicast Address Space page carries a section headed "Dynamic Multicast Group IDs", and its Reference block lists exactly two documents: "RFC 3307: Allocation Guidelines for IPv6 Multicast Addresses" and "RFC 10028: Updates to Dynamic IPv6 Multicast Address Group IDs". That is the pair Section 6 said it would list, in the order it named them, with their titles spelled out.

The section then prints two tables rather than one. The first has two columns, Range and Registration Procedures, and three rows: 0x90000000-0xEFFFFFFF, Standards Action; 0xFD000000-0xFDFFFFFF, Private Use; 0xFE000000-0xFEFFFFFF, Experimental Use. The second has three columns, Range, Description and Reference, and the same six ranges as Table 2 of the standard, with the same description text beside each one.

The plain-text export prints the same six rows and the same three procedure rows, with the RFC numbers bracketed throughout — "Defined in [RFC 2730], range assigned in [RFC 10028]" where the HTML page has no brackets. The CSV export has three columns and exactly six data rows, and the 0x90000000-0xEFFFFFFF row's Reference field is empty, which is what "Unassigned" would predict.

The comparison, row for row

This is where a reader holding both documents gets to work. Six ranges in the standard; six rows in the registry. The same notation: 32-bit values in hexadecimal, two hyphens and an en dash nowhere in sight. The same six descriptions, including the one that is simply the word "Unassigned". No range appears on one side and vanishes on the other.

My own count, run on the printed strings and labelled as mine rather than the documents': the six ranges are contiguous. Each begins one value above the previous range's end, the first starts at 0x80000000 and the last closes at 0xFFFFFFFF, and Section 6 sets 0x80000000 to 0xFFFFFFFF as the field in which any future range must sit. No gap, no overlap. A stranger can walk the whole band with a pencil.

One reference cell is a rewrite rather than a copy. Table 2's MADCAP reference reads "Defined in [RFC2730], range assigned in RFC 10028"; the registry serves it as "range assigned in RFC 10028" on the HTML page and "range assigned in [RFC 10028]" in the exports. The substance survives; the brackets do not.

The widest change, and who printed it

Table 1 gave MADCAP the server-allocation column across 0x80000000-0xFEFFFFFF. Table 2 confines MADCAP to 0x80000000-0x8FFFFFFF and hands 0x90000000-0xEFFFFFFF the description "Unassigned". That is a large subtraction, and the standard does not leave it to inference: Section 4 records that the MADCAP range is reduced, and adds that existing MADCAP deployments should use an updated implementation or operate without other IPv6 multicast address allocation protocols. The arithmetic and the printed sentence agree, which is the useful order for them to arrive in.

The field that does not close

Now the disagreement. RFC 10028 Section 6 prints one registration policy for the whole registry and names it: Standards Action. The registry prints a procedure table with three lines — Standards Action, Private Use, Experimental Use — and those three lines cover three of the six allocation rows. The other three rows, 0x80000000-0x8FFFFFFF, 0xF0000000-0xFCFFFFFF and 0xFF000000-0xFFFFFFFF, sit in a table whose columns are Range, Description and Reference, with no procedure printed against them anywhere in the section.

It would be easy to call this a contradiction and stop. The more accurate reading is narrower, and it matters for anyone who ever wants to request a range. The two extra labels are already in the standard's own Description column: "Reserved for Private Use" and "Reserved for Experimental Use". What the registry does is promote those two purpose labels into a procedure column, where they sit beside the one label Section 6 named. So the registry does not introduce facts the standard withheld.

It changes the shape of a field: one registry-wide policy sentence becomes three per-range lines, and the mapping from the other three rows to any of those lines is left for the reader.

Neither reviewed document ranks the other, and neither states how a per-range table and a single policy line are meant to be reconciled. Naming the seam is as far as the paper goes.

The second promise, in another registry

Section 6 makes a second commitment, about a different registry. It states that IANA updated the FF3X:0:0:0:0:0:8000:0-FF3X:0:0:0:0:0:FFFF:FFFF entry in "Unicast-based (Including SSM) Multicast Group IDs" so that it uses dynamic assignment per the protocols listed in the new registry, with the reference changed to RFC 10028. The registry shows that row with the description "This range uses dynamic assignment according to the protocols listed in the Dynamic Multicast Group IDs registry" and the reference RFC 10028. Second promise, kept.

Inside it sits a date that the reviewed pages leave alone. That row carries Date Registered 2026-03-19. RFC 10028 prints its publication as August 2026. A row that cites a standard as its reference is dated five months before the month that standard carries. This piece asserts no cause for the ordering, and the two documents assert none either.

One correction on the way in

The registry's name is the trap. "Multicast Address Space" covers two IANA registry groups, and the group-ID registry is not on the IPv4 one. The IPv4 page is headed "IPv4 Multicast Address Space", carries the Local Network Control Block 224.0.0.0/24 and the Internetwork Control Block 224.0.1.0/24 with RFC 5771 procedures, and contains no "Dynamic Multicast Group IDs" section and none of these group-ID ranges. A reader who searches for the registry by the words in its group name lands one page early and finds a registry that looks right and is not. The correction costs one click, and it is the kind of error a documentary comparison has to survive before it can say anything.

What these two documents cannot tell you

A registry row sets a range aside. It is not an allocation inside that range, and no allocation record was obtained in this review. A row is also not evidence that anything is deployed: nothing in either document states that a host or an application uses a group ID from any of the six ranges, and "Unassigned" measures nothing about demand. "Standards Action" is a rule about who may change the registry, not a record of a change already made.

The dates cannot be joined either. RFC 10028 prints a month. The registry section as captured prints no date of its own; a last-updated value of 6 November 2026 was reported for the group's plain-text export, whose captured excerpt begins at the Dynamic Multicast Group IDs heading rather than at the header lines. A month and a day cannot be subtracted, so the two printed values stand apart and no interval is drawn between them.

Five things were not obtained in this review and would each move the argument: the registry's change log, which could date every edit; any request or approval record under Standards Action for a group-ID range; the IANA Functions Contract or any instrument reconciling a per-range procedure table with a single policy line; the PIM working-group record behind RFC 10028; and RFC 3307 itself, beyond its appearance in Table 1's caption and the registry's reference block.

So this is the state of the promise, as of 22 November 2026. The rows are kept to the letter, and a reader can verify that in an afternoon with a pencil. The policy is kept in a different shape than the one the standard described, and nothing in either document says which shape controls. IANA will publish the next edit to that registry as a row; whoever wants to know what the edit means will have to reopen RFC 10028 as well, because the two publications are one instrument split across two publishers and neither is complete on its own.