Hi Adrian, Sorry for the late reply. Please see my answers below inline.
Thanks, Yingzhen On Wed, Nov 6, 2024 at 3:38 AM Adrian Farrel <adr...@olddog.co.uk> wrote: > Hey, Yingzhen. > > [snip] > > I am happy with the concept of providing an initial discussion venue for > such topics, but: > > - What does “incubation” mean in practice? I think this needs to be > spelled out in the charter text because, as it stands, it is unclear where > you would draw the line. Why would this not result in tens of documents > being pushed to RFC on the topic of (for example) wet-string routing? How > would the WG handle requests to discuss 12 new I-Ds on a new topic at an > IETF? How would the WG protect the other chartered work of the working > group against being swamped? > I’d suggest that “particularly focusing on problem statements” be made > more limiting such as, “by developing and discussing problem statement and > requirements documents”, and that “prior to achieving consensus” be > end-limited as “prior to determining whether there is consensus or not for > the creation of a new working group.” > > [Yingzhen]: > > The "incubation" means to provide a venue for initial discussion > which RTGWG has been doing already. Ideally we may give a new proposal/work > one or two opportunities to present, and based on the feedback of the > community we'll decide whether to continue. Use your wet-string routing > example, if we really get tens of documents on the topic, and assuming > they're not from the same author, I'd say most likely the community is > interested in the topic. We should help to define the problem scope and > recommend for a BoF before there are tens of documents on this topic. > > As for agenda building, we will always put WG documents on higher > priority. Meanwhile we can always make use of interims and side meetings. > I'm actually not too worried about getting too much work. It's way better > than having nothing to work on. > > Regarding the charter update, I merged your suggestion. So how about > something like: > > Incubating new routing-related technologies by developing and discussing > problem statements and requirements documents, prior to determining whether > there is consensus or not for the creation of a new working group. Initial > topics include, but are not limited to, satellite routing, data center > routing, and networking for AI clusters. > > > > [AF] This is all good stuff, and I don’t see anything I disagree with > substantially. > > But… :-) > > The way to think about charter text is not just about how to guide the > current (wise and competent) chairs, but how to guide the future (evil or > dumb) chairs, and how to guide the highly-excitable and inventive WG > participants. > > So, probably, we just need to cook some words that not rely on a single > key word, but sets the scope and process and priority. > [Yingzhen]: thanks for this "But". This is what we need. I agree with you that the charter of a WG should be specific, but not so specific that it may go out of date in a week of recharter. > > - The text “This includes, but is not limited to” is, I think intended > to say “The initial list of candidate topics is,” with an addition of > “other topics may be added after discussion with the WG chairs”. But… > > > - What does this initial list actually add? > - Will you track those other topics? How do the chairs decide? What > happens when an idea won’t go away, or keeps coming back? > > [Yingzhen]: > > Are you suggesting we shouldn't include this initial list? The "not > limited to" meant to say we may also work on other topics. We can't predict > the topics that may pop up, and recharter does have some overhead and take > some time. Of course, if there is a topic that the community thinks RTGWG > should work on, we can always recharter to include it. > > > > [AF] If you are saying that further topics will be added to the charter > before they can be discussed (I don’t think you are saying that) then > having the list would be fine. > > But otherwise, I think you are just making a list that will be out of date > by the end of the week. > > So, how about, “A list of topics currently under consideration by the WG > for incubation will be maintained on the WG Wiki page”? > > > [Yingzhen]: A new topic may pop up in RTGWG any time, so it's not possible for us to make a complete list of topics that the WG will be working on. How about if the WG agrees to work on the problem statement of a topic, for example the problem statement of wet-string routing, we will add this to the milestones of the WG. How about the following text for charter update: *Incubating new routing-related technologies by developing and discussing problem statements and requirements documents, prior to determining whether there is consensus or not for the creation of a new working group. If the working group agrees to pursue a problem statement or requirements document, it will be added to the group's milestones.* > Since we just had the perceptive/adaptive routing side meeting, and since > we have the AIDC mailing list, would you imagine that the day after > rechartering all of that work would immediately move into RTGWG? > > [Yingzhen]: > > The preceptive/adaptive routing side meeting is actually a good example to > show that efforts will also be made outside of RTGWG to figure out the > problem space and whether there is something that the IETF can work on. > Same for the AIDC side meetings and the mailing list where we also share > the latest technology developments in the industry. > > > > [AF] Right. Two things should be clear. > > 1. New topics can be spun up in the RTG Area without “incubating” them > in RTGWG. Thus, side meetings, I-Ds, BoFs, WG formation do not need to use > RTGWG. > 2. RTGWG is available as **an** option for airing new topics. > > However, in this particular case, when stuff has advanced on several > fronts, I think there is a search for coordination advice…. > - If perceptive/adaptive routing is already under RTGWG incubation, it > would be helpful to know > > - If it is an option for the perceptive/adaptive routing proponents to > incubate in RTGWG, then they need to self-organise to that end > > - If the AIDC list is the place to self-organise for perceptive/adaptive > routing, this is **highly** non-obvious from the name of the list :-) > [Yingzhen]: There are already drafts in RTGWG about different aspects of networking technologies in data centers for LLM training. I'm not sure whether you consider them related to perceptive/adaptive routing. These drafts are still at early stages and need further work. As for the AIDC mailing list, it was created to facilitate discussions related to the AIDC side meetings, where we discuss new technologies in DCs for AI. I thought the name was obvious, and it seems I'm mistaken. :) > > > Cheers, > > Adrian >
_______________________________________________ rtgwg mailing list -- rtgwg@ietf.org To unsubscribe send an email to rtgwg-le...@ietf.org
