Summary
- RFC 9119, published in October 2021 in the Informational category, says well-known multicast issues "have prevented the deployment of multicast in 802.11 (Wi-Fi) and other local-area wireless environments", and its section 3.1.1 puts the mechanism at the radio link: no acknowledgements, therefore no retransmission.
- Revision 10 of draft-ietf-pim-multicast-lessons-learned, dated 1 November 2026, puts the mechanism in the architecture — every Layer 3 hop must support a multicast routing protocol — and prints its own instruction that it is inappropriate to cite the document "other than as work in progress".
- M. McBride of Futurewei is the only name printed on both author lines, and the two documents disagree about what their own status is worth: one carries the IETF consensus sentence, while the other sits on a record page that prints "Intended RFC status (None)" beside a cover page printing "Intended Status: Informational".
Two author lines, five years apart
The first page is an RFC Editor front matter, and it states its own terms before it states anything else: internet engineering task force stream, RFC 9119, Informational category, published October 2021, ISSN 2070-1721 (RFC 9119). Under those lines sit five names and five affiliations, in this order: C. Perkins of Lupin Lodge, M. McBride of Futurewei, D. Stanley of HPE, W. Kumari of Google, JC. Zúñiga of SIGFOX.
The second page is an Internet-Draft cover, and its terms are different in kind. Workgroup: Network Working Group. Internet-Draft: draft-ietf-pim-multicast-lessons-learned-10. Published: 1 November 2026. Intended Status: Informational. Expires: 5 May 2027 (the draft's front matter). Four names follow: D. Farinacci of lispers.net, L. Giuliano of HPE, M. McBride of Futurewei, N. Warnke of Deutsche Telekom.
Put the two lists side by side and something narrow and load-bearing appears. Perkins is on the 2021 document only. Farinacci, Giuliano and Warnke are on the 2026 document only. One name survives the comparison: Mike McBride, listed at Futurewei on both pages. That is the whole reason this article can be written about a person rather than about a working group. Replace McBride with any other printed author on either list and the sentence collapses — you would be comparing one document to itself.
What that name does not prove is worth saying plainly, because the temptation to inflate it is strong. An author line records authorship, not authorship share. Neither page says who wrote which paragraph, and nothing in this article claims that McBride drafted section 3.1.1 of the 2021 document or section 3.9 of the 2026 draft. What the record shows is smaller and more durable: he is the one person whose name the two documents share, which means the five-year argument between them is an argument inside one career, not a dispute between strangers.
What the approved document puts at the radio link
The 2021 document is about wireless, and it says so in its first sentence of substance. Its abstract runs: "Well-known issues with multicast have prevented the deployment of multicast in 802.11 (Wi-Fi) and other local-area wireless environments" (RFC 9119). The scope is deliberately modest — the known limitations of wireless, primarily 802.11, Layer 2 multicast, plus the enhancement features the IETF and IEEE 802 have specified and some operational choices that can improve performance.
The mechanism appears in section 3.1.1, titled Multicast Reliability, and it is a link-layer story told in five steps. Multicast traffic is typically much less reliable than unicast traffic. Because multicast is point-to-multipoint, guaranteeing reception at all recipients would take multiple acknowledgements. There are no ACKs for multicast packets, so the access point cannot know whether a retransmission is needed. The sender never backs off.
Therefore, as the section puts it, "It is not uncommon for there to be a packet loss rate of 5% or more, which is particularly troublesome for video and other environments where high data rates and high reliability are required."
That is a diagnosis with an address. The failure lives on the radio hop between an access point and a device, and it lives there because of a protocol property — no acknowledgement — rather than because of anyone's procurement decision. Read alone, section 3.1.1 explains why streaming video to a lecture hall of laptops went badly. It does not explain why thirty years of multicast routing protocols did not reach the public Internet, and it does not try to.
The heading inventory shows how tightly the document is bounded to that job: eleven numbered sections, from Introduction through Identified Multicast Issues, Multicast Protocol Optimizations, Operational Optimizations, Multicast Considerations for Other Wireless Media, Recommendations and Ongoing Discussion Items to Informative References, with the optimizations running through Proxy ARP, IPv6 address registration, buffering, hardware queue depth, layer-2-to-unicast conversion, Directed Multicast Service, Automatic Multicast Tunneling and GroupCast with Retries (RFC 9119 table of contents). A reader looking for a verdict on multicast in the global Internet will not find one. The document is a workshop manual for the last hop, and it knows it.
What the working document puts in the architecture
Five years later, the second document goes where the first one did not. Its Introduction opens in the 1980s with Steve Deering's service model — packets from a group are received only if a receiver explicitly requests them — then lists the protocols built on top of that model across IPv4, IPv6, tunnels and label-based solutions: DVMRP, PIM-DM, PIM-SM, PIM-BIDIR, PIM-SSM, MSDP, MBGP, MVPN, P2MP RSVP-TE, MLDP, BIER, LISP, MOSPF, IGMP, MLD, "and several others" (the draft's Introduction). Then it names the mood of the last three decades in one sentence: "there has been much angst over deploying IP Multicast over the last 30 years."
The authors say why they wrote the document at all, in the final two sentences of that Introduction: despite multicast-related best current practices, they "felt it important to have a document which helps answer some of these questions through identifying the lessons learned from multicast development and deployment over the last 30 years." The lesson inventory is the document's own section list — twenty-three numbered entries, from 3.1 DVMRP and 3.2 MOSPF through 3.9 All or Nothing Problem, 3.11 Network Based Source Discovery and 3.14 Premature Optimization, down to 3.22 Mobility and 3.23 Multicast OAM (the draft's table of contents).
Two of those twenty-three carry the argument that matters here.
Section 3.9 begins with a requirement stated flatly: "For multicast to function, every layer 3 hop between the sourcing and receiving end hosts must support a multicast routing protocol" (section 3.9). The authors concede that enterprises and walled-garden networks can absorb that, because the perceived benefits — financial, video distribution, service-provider MVPN — exceed the cost of deploying. Then they turn to the global Internet, where the cost-benefit case for any service is not likely to be universally agreed, and conclude that this all-or-nothing requirement "tends to create an insurmountable barrier". Their generalisation is blunt: "any technology that requires new protocols to be enabled on every interface on every router and firewall on the Internet is not likely to succeed." The section also says IPv6 suffers the same challenge, which it offers as an explanation for why IPv6 "has not been ubiquitously deployed across the Internet to the same degree as IPv4, despite decades of trying" — the draft's claim, stated as the draft's claim. Overlay networking, AMT and LISP are named as ways to buy incremental deployment instead.
Section 3.11 moves to source discovery. In ASM, "the network is responsible for discovering all multicast sources", and the authors say that responsibility "leads to massive protocol complexity, which imposes a huge operational cost for designing, operating and troubleshooting multicast" (section 3.11). In SSM, discovery moves out of the network to an out-of-band mechanism, typically the application layer, and that removal eliminates shared trees, PIM register encapsulation and decapsulation, rendezvous points, SPT switchover, data-driven state creation and MSDP. The result, they write, is that "PIM-SSM is merely a small subset of PIM-SM functionality." Then the sentence that gives the section its weight: "The key insight is that source discovery is not a function the network should provide." Their analogy is a database of every active website that OSPF, IS-IS and BGP are never asked to keep, and their heuristic has a name — what would unicast do.
The diagnosis moved; it did not shrink
Set the two mechanisms beside each other and the shape of the change is visible without embellishment. In 2021, multicast's trouble is located at a link: no acknowledgements, no retransmission, loss on wireless hops. In 2026, the trouble is located in a requirement: every hop must speak a multicast routing protocol, and every router and firewall on the path must bend to it. The first explanation is a property of radios.
The second is a property of the Internet's ownership structure — which is why the same authors can say that a technology needing universal enablement "is not likely to succeed" without needing to measure anything.
There is a temptation to call the second account the corrected one. Resist it. These are not competing measurements of the same quantity; they are diagnoses at different layers, written for different purposes, by overlapping hands. RFC 9119 exists to describe wireless limitations and recommend operational choices that mitigate them. The draft exists to explain a thirty-year deployment record and to argue about where certain functions belong. A reader can accept both without choosing, and the authors never present them as rivals.
What has genuinely broadened is scope, and the difference in length reflects it: an eleven-section report about the last hop versus a document whose section 3 alone carries twenty-three retrospectives across four decades of protocol design. If the 2021 paper explains why the wireless edge of multicast is expensive, the 2026 draft explains why the middle of the network never agreed to carry it in the first place.
The 5% figure is the RFC's sentence, not a survey of the Internet
One number in this pair travels badly, so it needs a fence around it. The RFC's section 3.1.1 states it is not uncommon for there to be a packet loss rate of 5% or more on wireless multicast. That is the RFC making a statement about packet loss rates in the environment it studies. It is not a measurement taken for this article, it is not a rate observed on any particular network, and no arithmetic in this piece adds to it, averages it or extrapolates from it.
Treating a document's characterisation of a common condition as though it were a traffic study is the easiest available error here, and it would convert a link-layer explanation into a claim about the entire Internet that nothing in either document supports.
The same discipline applies to the draft. Its ratios are verbal — "massive protocol complexity", "a huge operational cost", "an insurmountable barrier" — and its twenty-three lessons are engineering judgement accumulated over decades, not results tabulated from a corpus. Neither document reports any measurement of running multicast traffic. That absence is the single most important limitation on the pair, because it means the disagreement about where multicast failed is an argument about architecture, and architecture is exactly the kind of thing reasonable engineers can dispute without new data.
A printed seam: (None) beside Informational
The most arresting detail in this pair is not in either body. It is in the metadata.
The draft's cover page prints an intended status of Informational. The Datatracker document record for the same document prints two lines: "Intended RFC status (None)" and "IESG state I-D Exists" (the document record). The same record describes the document as an active Internet-Draft in the pim working group, on the IETF stream.
Read that as a seam in the machinery, not a scandal. One surface tracks process state, and no IESG state has been reached, so it carries no intended status. The other surface is the document's own self-description, which is the authors' aspiration rather than a decision by anyone with the power to confer status. The cover page says what the authors intend; the record page says where the process actually stands. Both are accurate about their own subject. Neither is a mistake, and no page read for this article states why the record prints (None) — so this article does not supply a reason.
The consequence, though, is concrete and it is the second half of the story. Multicast's broadest published diagnosis currently exists in a document class whose own Status of This Memo says it is inappropriate to cite it "other than as work in progress", and which expires on 5 May 2027. The narrower diagnosis, the one about radios, is the one that carries the sentence every citation needs: "It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG)" — printed in the same memo that also says, in the sentence before, that the document "is not an Internet Standards Track specification; it is published for informational purposes" (RFC 9119).
Both sentences sit in the same paragraph on the same page. That is worth pausing on, because it is the ordinary condition of most useful technical publication: a document can be the product of community consensus and IESG approval while carrying no normative force whatsoever. A status line authorises scope. It does not certify that the diagnosis inside the document is true, and here nobody claims otherwise.
Why one name carries the comparison
The swap test is simple and it is the test that decides whether this is a story about a person or a story about an institution. Replace M. McBride on the 2021 author line with D. Stanley, W. Kumari or JC. Zúñiga, and the 2026 document has no author in common with it. Replace him on the 2026 cover with D. Farinacci, L. Giuliano or N. Warnke, and the 2021 document has no author in common with it. The comparison exists only because of the one name that appears twice, at the same affiliation, five years apart.
That is what makes the finding person-level rather than institutional. A working group can adopt a document; the IESG can approve one; an RFC Editor can publish one. None of those bodies can be the through-line between a wireless problem statement approved in 2021 and an architectural post-mortem still in draft in 2026. A named author can be exactly that, and one is.
What the pair actually establishes
Multicast's non-deployment has a written diagnosis, and it was written twice by overlapping hands at two different addresses. The first version, published and approved, is narrow: radios do not acknowledge multicast packets, so nothing gets retransmitted, so loss is high and video suffers. The second version, unfinished and explicitly not yet citable, is general: the technology demands support on every Layer 3 hop and every interface of every router and firewall, and it asks the network to perform source discovery that the network should not perform.
The order of publication is the uncomfortable part. The general explanation is the one you are told not to rely on yet. The citable explanation is the one that only covers the last hop. Anyone who wants to quote a strong claim about why multicast failed across the Internet as a whole must either quote a document that instructs them to call it work in progress, or wait — and the document they would be waiting on expires on 5 May 2027, with its record page printing no intended status at all.
That is not a failure of the IETF, and this article does not present it as one. It is what a mature standards process looks like from the inside: the cautious, narrower claim gets the consensus sentence, because it is the claim a community can agree on, while the sweeping claim stays in draft for as long as it takes. The cost is borne by readers who need the general account now, and by the author whose name connects the two.
The limits of what was read
Two boundaries belong beside the argument rather than in a footnote. The draft's rendered text was available up to section 3.3 and truncates there, so sections 3.1, 3.2, 3.4 through 3.8, 3.10 and 3.12 through 3.23 were not read in their own words; their headings come from the table of contents, and no claim here rests on their contents. The RFC's rendered text truncates in section 3.1.2, so its complete heading list comes from extracted PDF text rather than a line-by-line inspection of the layout.
What is left out is more important than what is in. Neither document reports traffic measurements. Neither document states why a record page prints (None). No page read here describes how each author's share of the writing was divided, and this article makes no attempt to guess. Those are the edges of the evidence, and the finding sits comfortably inside them: one name, two documents, two addresses for the same failure, and a status line that decides which version a careful reader is allowed to quote today.
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






