[{"author": "Ingemar Johansson", "text": "<p>Good afternoon</p>", "time": "2024-11-08T13:02:01Z"}, {"author": "Simone Ferlin", "text": "<p>Hello! Welcome Ingemar (and everyone)</p>", "time": "2024-11-08T13:02:24Z"}, {"author": "Gorry Fairhurst", "text": "<p>Great Agenda this meeting!</p>", "time": "2024-11-08T13:22:33Z"}, {"author": "Simone Ferlin", "text": "<p>Thanks Gorry :)</p>", "time": "2024-11-08T13:26:47Z"}, {"author": "Jonathan Morton", "text": "<p>I don't think even BBRv3 has actual ECN support yet.</p>", "time": "2024-11-08T13:28:53Z"}, {"author": "Jonathan Morton", "text": "<p>So switching it on for a BBR flow just stops the AQM from signalling using loss.</p>", "time": "2024-11-08T13:29:39Z"}, {"author": "Ingemar Johansson", "text": "<p>As I understood it, BBRv3 is more L4S support. This can explain that Cubic is pushed down</p>", "time": "2024-11-08T13:29:44Z"}, {"author": "Jonathan Morton", "text": "<p>I think they were still trying to decide whether to do L4S-style or RFC-3168 compliant ECN support.</p>", "time": "2024-11-08T13:30:17Z"}, {"author": "Ingemar Johansson", "text": "<p>But perhaps L4S was disabled in BBRv3, in BBRv2 there was an RTT constrain for L4S</p>", "time": "2024-11-08T13:30:34Z"}, {"author": "Gorry Fairhurst", "text": "<p>This week, some thought that ECT(0) response might be a step to enabling a first step in BBRv3... perhaps.</p>", "time": "2024-11-08T13:31:16Z"}, {"author": "Jason Livingood", "text": "<p>Great paper and presentation! <span aria-label=\"clap\" class=\"emoji emoji-1f44f\" role=\"img\" title=\"clap\">:clap:</span></p>", "time": "2024-11-08T13:31:33Z"}, {"author": "Joseph Beshay", "text": "<p>Re BBRv3+ECN discussion. It's here: <a href=\"https://github.com/ietf-wg-ccwg/draft-ietf-ccwg-bbr/issues/7\">https://github.com/ietf-wg-ccwg/draft-ietf-ccwg-bbr/issues/7</a></p>", "time": "2024-11-08T13:32:03Z"}, {"author": "Jonathan Morton", "text": "<p>I made a proposal on the issue open</p>", "time": "2024-11-08T13:32:09Z"}, {"author": "Jonathan Morton", "text": "<p>which is basically that an RFC-3168 compliant response should be easy to implement and probably sufficient.</p>", "time": "2024-11-08T13:32:37Z"}, {"author": "Seifeddine Fathalli", "text": "<p>I agree that the 15s gap is a corner case, nevertheless in  the Internet we can't guarantee that won't happen.</p>", "time": "2024-11-08T13:34:04Z"}, {"author": "Simone Ferlin", "text": "<p>The URL of the work : <a href=\"https://inet-bbrv3eval.mpi-inf.mpg.de/\">https://inet-bbrv3eval.mpi-inf.mpg.de/</a></p>", "time": "2024-11-08T13:34:39Z"}, {"author": "Peter Heist", "text": "<p>arguably, high RTT flows getting higher throughput may be the opposite of what is desirable, from the standpoint that upon a rate drop, the latency harm to short RTT flows will be that much higher as the high RTT flow takes longer to react (and if it also has higher throughput, that doesn't help)</p>", "time": "2024-11-08T13:44:06Z"}, {"author": "Peter Heist", "text": "<p>thanks for this work!</p>", "time": "2024-11-08T13:44:13Z"}, {"author": "Jonathan Morton", "text": "<p>Actually, a JFI of 0.5 could be COMPLETELY UNACCEPTABLE in a two-flow contest, because it represents one of the flows getting zero throughput.</p>", "time": "2024-11-08T13:44:58Z"}, {"author": "Jonathan Morton", "text": "<p>I would invite consideration of the \"harm index\", which is easier to interpret, rather than JFI.</p>", "time": "2024-11-08T13:45:58Z"}, {"author": "Randell Jesup", "text": "<p>Good work also.   I am concerned about the convergence delay, especially when used for realtime media (such as over QUIC/WebTransport, etc).   Also, I think it would very interesting to see how BBR affects WebRTC realtime CC flows - if it kills them (compared to Cubic perhaps), or seriously dominates them, that's problematic</p>", "time": "2024-11-08T13:47:13Z"}, {"author": "Emilia Weyulu", "text": "<p>@Jonathan yes, 0.5 JFI is not desirable.</p>", "time": "2024-11-08T14:00:23Z"}, {"author": "Gorry Fairhurst", "text": "<p>Would love to know how this is impacted by Hystart++ rather than Hystart...</p>", "time": "2024-11-08T14:02:08Z"}, {"author": "Emilia Weyulu", "text": "<p>with regards to harm, I agree re explainability. But I don\u2019t think the insights about the CCA\u2019s performance will change by looking at harm instead of JFI</p>", "time": "2024-11-08T14:04:03Z"}, {"author": "Jonathan Morton", "text": "<p>I think \"cwnd*\" is essentially a standin for the steady-state BDP.</p>", "time": "2024-11-08T14:04:14Z"}, {"author": "Jonathan Morton", "text": "<p>So it is not known in advance, but is determined a-posteriori.</p>", "time": "2024-11-08T14:04:29Z"}, {"author": "Lars Eggert", "text": "<p>Not that's a SIGCOMM-y title if I ever saw one</p>", "time": "2024-11-08T14:05:17Z"}, {"author": "Lars Eggert", "text": "<p><em>Now</em></p>", "time": "2024-11-08T14:05:24Z"}, {"author": "Jonathan Morton", "text": "<p>The \"grow faster this period\" criterion seems to leverage the same measurements that HyStart collects, and just uses them for a different purpose.</p>", "time": "2024-11-08T14:05:48Z"}, {"author": "Jonathan Morton", "text": "<p>The distinction between the ack-clocked and paced sending periods in each RTT seems to be instructive.</p>", "time": "2024-11-08T14:06:34Z"}, {"author": "Jonathan Morton", "text": "<p>If the ack-clocked segment of data is acked over a relatively large fraction of the BDP, you can infer that the delivery rate is limited by the path bottleneck, and that indicates that you are approaching the path capacity.</p>", "time": "2024-11-08T14:07:58Z"}, {"author": "Chris Box", "text": "<p>Isn't all this assuming that most congestion occurs in the middle of the network where very large pipes meet? My impression is that's not true, because these pipes are usually dimensioned to be larger than demand. I expect most congestion is near the lowest-bandwidth parts of the route.</p>", "time": "2024-11-08T14:15:06Z"}, {"author": "Jonathan Morton", "text": "<p>Yes, that's usually true (or at least planned to be true).</p>", "time": "2024-11-08T14:15:34Z"}, {"author": "Ingemar Johansson", "text": "<p>Yes, looks quite similar to ConEx</p>", "time": "2024-11-08T14:23:44Z"}, {"author": "Ingemar Johansson", "text": "<p>Agree, Bob is indeed missed</p>", "time": "2024-11-08T14:26:43Z"}, {"author": "Simone Ferlin", "text": "<p>Keep going and I fix my audio.</p>", "time": "2024-11-08T14:29:55Z"}, {"author": "Akshay Narayan", "text": "<p>@chris, no, we don't make that assumption about where the congestion is. the congestion's location will affect the appropriate mechanism because pipes are differently sized, but the bandwidth allocation framework is mechanism-agnostic</p>", "time": "2024-11-08T14:34:36Z"}, {"author": "Chris Box", "text": "<p>LTE uses a 1ms timeslot</p>", "time": "2024-11-08T14:36:59Z"}, {"author": "Chris Box", "text": "<p>Surely the problem is that the goal for wifi has been to achieve higher throughputs, as if that is the only thing that matters. Latency needs more focus.</p>", "time": "2024-11-08T14:49:07Z"}, {"author": "Jason Livingood", "text": "<p>+1 chris</p>", "time": "2024-11-08T14:50:32Z"}, {"author": "Ingemar Johansson", "text": "<p>Just for the record, really appreciated to get the opportunity to bring this up. Thanks chairs !</p>", "time": "2024-11-08T14:50:36Z"}, {"author": "Reese Enghardt", "text": "<p>For sure, thank you for participating! Props to Simone for proposing the panel discussion</p>", "time": "2024-11-08T14:51:13Z"}, {"author": "Simone Ferlin", "text": "<p>Thanks for accepting the invite. When we need to understand congestion in mobile networks, we know who is the reliable reference</p>", "time": "2024-11-08T14:51:38Z"}, {"author": "Jason Livingood", "text": "<p>Maybe we need WiFi slicing. <span aria-label=\"rolling on the floor laughing\" class=\"emoji emoji-1f923\" role=\"img\" title=\"rolling on the floor laughing\">:rolling_on_the_floor_laughing:</span></p>", "time": "2024-11-08T14:51:56Z"}, {"author": "Michael Welzl", "text": "<p>This discussion is really, really interesting. Kudos to the chairs for organizing this.</p>", "time": "2024-11-08T14:52:57Z"}, {"author": "Simone Ferlin", "text": "<p><span aria-label=\"woman bowing\" class=\"emoji emoji-1f647-200d-2640\" role=\"img\" title=\"woman bowing\">:woman_bowing:</span></p>", "time": "2024-11-08T14:54:01Z"}, {"author": "Jason Livingood", "text": "<p>More seriously: seems we need better TSVWG-IEEE 802.11 (or WiFi Alliance) collaboration. Optimally a way to use different WMMs or EDCAs for different ECN or DSCP marking. The mappings from IETF-land to IEEE-land are dated and not suitable to today's internet.</p>", "time": "2024-11-08T14:54:18Z"}, {"author": "Lars Eggert", "text": "<p>PILC = <a href=\"https://datatracker.ietf.org/wg/pilc/about/\">https://datatracker.ietf.org/wg/pilc/about/</a></p>", "time": "2024-11-08T14:55:32Z"}, {"author": "Ingemar Johansson", "text": "<p>As we are all involved in making something good better, it is important to know that things don't break down if we do a little wrong, it is just a bit suboptimal</p>", "time": "2024-11-08T14:55:40Z"}, {"author": "Jason Livingood", "text": "<p>Good point @Bob -- in the post-gigabit world, the main issue is on the wireless LAN.</p>", "time": "2024-11-08T14:56:01Z"}, {"author": "Simone Ferlin", "text": "<p>Shouldn't something like the topic of this discussion be documented more formally?</p>", "time": "2024-11-08T14:58:35Z"}, {"author": "Simone Ferlin", "text": "<p>I think it is a lot of good points here that are at risk of being lost post-meeting</p>", "time": "2024-11-08T14:58:48Z"}, {"author": "Reese Enghardt", "text": "<p>Right, we'll have the meeting recording, the slides, and the notes, but maybe there is room for something more persistent</p>", "time": "2024-11-08T14:59:27Z"}, {"author": "Reese Enghardt", "text": "<p>I remember in CCWG, there was interest in different link types and their \"descriptions\"</p>", "time": "2024-11-08T15:00:01Z"}, {"author": "Spencer Dawkins", "text": "<p>So, we're waiting until the afternoon of Friday to talk about PILC? I need to be more awake for that! (speaking as the  former PILC co-chair)</p>", "time": "2024-11-08T15:00:05Z"}, {"author": "Simone Ferlin", "text": "<p>This, Reese ^^^</p>", "time": "2024-11-08T15:00:14Z"}, {"author": "Lars Eggert", "text": "<p>I think it might be worthwhile to do PILCv2. A lot of people are unaware of that work, and it is still very relevant, albeit in need of an update.</p>", "time": "2024-11-08T15:00:41Z"}, {"author": "Spencer Dawkins", "text": "<p>Lars, yes, and yes on need for an update (the first PILC doc was about 2.5G links ....)</p>", "time": "2024-11-08T15:01:49Z"}, {"author": "Simone Ferlin", "text": "<p>If this moves forward in CCWG, it could be somethinf ICCRG can contribute to.</p>", "time": "2024-11-08T15:01:49Z"}, {"author": "Reese Enghardt", "text": "<p>Definitely, yes</p>", "time": "2024-11-08T15:02:01Z"}, {"author": "Lars Eggert", "text": "<p>In the same space as PILC, we did try to do a TERNLI WG, which failed to charter: <a href=\"https://mailarchive.ietf.org/arch/msg/ternli/_wlpZQNZjyhtGQzIKbBrfYvyBVk/\">https://mailarchive.ietf.org/arch/msg/ternli/_wlpZQNZjyhtGQzIKbBrfYvyBVk/</a></p>", "time": "2024-11-08T15:02:42Z"}, {"author": "Reese Enghardt", "text": "<p>Interesting, I hadn't heard of TERNLI. I think these discussions are appropriate for ICCRG, maybe we can get away without chartering</p>", "time": "2024-11-08T15:03:27Z"}, {"author": "Lars Eggert", "text": "<p>TERNLI mostly failed because three of the main proponents became ADs together and lost all time to spend on it :-)</p>", "time": "2024-11-08T15:04:20Z"}, {"author": "Jonathan Morton", "text": "<p>Adding CoDel to the WiFi stack actually turns out to be a natural fit.</p>", "time": "2024-11-08T15:07:33Z"}, {"author": "Jonathan Morton", "text": "<p>See the \"make-wifi-fast\" project\u2026</p>", "time": "2024-11-08T15:07:55Z"}]