IETF 123 Madrid
MBONED Agenda
Wed, Jul 23, 2025
09:30 - 11:00 CEST
Wed Session I, Hidalgo
Chat Log:
https://zulip.ietf.org/#narrow/stream/101-mboned/topic/ietf-123
Video log: https://www.youtube.com/watch?v=36Y9uT_E5t8
Full session recording (incl transcript, polls, chat, video,etc):
https://www.meetecho.com/ietf123/recordings#MBONED
https://meetecho-player.ietf.org/playout/?session=IETF121-MBONED-20241106-1500
Note taker: Sandy Zhang
Status of WG items
Status of WG items
Chairs, 10 min
Recharter and updated milestones since 122
Sandy Zhang: multicast YANG and redundant ingress failover drafts
addressed all the comments received.
Yisong Liu: AMT YANG requested yang doctors review and hasn't
received comments.
Multicast usage in LLM MoE (draft-zhang-rtgwg-llmmoe-multicast)
Sandy Zheng, 15 min
Mike: do you think about the scale problem?
Sandy: some of the multicast requirements are just inside the node,
for MoE there is explicit requirements across leaf/spine switch. For
the specific use case of MoE, just 9 (for now) experts are selected
at a time. And from the experience of cloud operator, the number of
experts will not exceed 100. When we discuss the multicast use cases
to send data to thousand of GPUs, it's better to have the explicit
use cases and requirements.
Hitoshi: Interesting and positive topic. Which working group does
this draft want to be proceed.
Sandy: For the use cases MBONED might be the suitable WG. We also
have slots in BIER, RIFT session.
Lenny: is this potential solution just for MoE and LLM use case or
for all the AIDC training clusters?
Sandy: There are existed broadcast and multicast operations in LLM.
But they may just inside the node and be solved by private multicast
solutions or by unicast only. The MoE use case requires the data
transfer across leaf/spine switch, it's time to use network layer
multicast technologies to help with it.
Max: Many papers for using multicast in LLM may be help. The
modification will also require the modification of LLM. For RDMA,
the data transfer function may be changed.
Sandy: Yes. The modification would be related to LLM coding and the
function of the data transfering.
Discovery Of Restconf Metadata for Source-specific multicast
(draft-ietf-mboned-dorms)
"As an individual: I think that it would be really valuable to see if
the IETF could progress CBAC as one way to introduce helpful tools for
coexistance of multicast and unicast traffic !!!
As WIT AD: I think the CC/circuit breaker part of the "problem" falls
within the expertise in CCWG. I'd encourage you to reach out to CCWG and
think of not only presenting your work, but in discussing the results in
test scenarios and the traffic sharing properties. CCWG has published
RFC 9743, and I am hopeful that the evaluation tests in this draft will
be really helpful in growing confidence that multicast traffic is safe,
even when sharing constrained paths. I expect there likely can be
stronger requirements to define how a CBAC receiver works (too much
currently is SHOULD and not specified). This will also let the community
justify publication as a PS.
Overall, thanks for working on this!"
Max: yes, we plan to go to CCWG in next IETF meeting. Wanted to update
doc first.
Mohamed Boucadair (from chat): in addition to what Gorry mentioned,
looking at BCP133 would be helpful.
Bandwidth-Aware Multicast
(https://pourdamghani.net/_pages/IFIP25.pdf)
Max Franke, 5 min
Lenny: P2MP RSVP is able to do bw reservation for multicast trees
today
Jeffrey: yes. the P2MP RSVP is the only way currently and the
simplest way to reserve bandwidth for multicast. The work may be in
TEAS WG.
Security and Privacy Considerations for Multicast Transports
(draft-krose-multicast-security)
Kyle Rose, 10 min
Lenny: are you looking to adopt this draft in MBONED WG?
Kyle: yes. and may get some attention from SEC area.
Omar: Long time on this work. The work in CDN and with the
unmodified applications would like to use in CDN, home server, etc.
Glad to see it.
Reading vote: yes 8, no 8
Hitoshi: secure IGMP/MLD included?
Kyle: this draft is not about the specifiction. It's more for
analysis. How to meet the security requiments. More about web
stream.
Hooman: make sure the security is from the source to the browser.
Kyle: it's not hop by hop. For anyone authority in the network who
can subscribe the demand
Lenny: may need assistance from SEC area.
Adoption vote: yes 14, no 3, no opinion 2
Flexicast QUIC (draft-navarre-quic-flexicast)
Louis Navarre, 10 min
Francois Michel: do you want to get adoption? Would be happy to help
moving this work forward.
Louis: happy to show the implementation and happy to move forward
this work.
Jeffrey: how many reveivers can this handle?
Louis: the test is 1000 receivers but it can be more.
Experiences with LISP Multicast deployments
(draft-vgovindan-lisp-multicast-deploy)
Prasad Govindan, 5 min
run out of time, no presentation this time.
Lenny: why adopted in LISP and not MBONED?
Prasad: More LISP specific, so expertise is there. But welcomes
MBONED input.