IETF
iccrg
iccrg@jabber.ietf.org
Tuesday, March 9, 2021< ^ >
jishac has set the subject to: Internet Congestion Control Research Group
Room Configuration
Room Occupants

GMT+0
[08:06:54] Simon Leinen joins the room
[11:50:03] Paolo Saviano_web_144 joins the room
[11:50:03] Juhamatti Kuusisaari_web_413 joins the room
[11:50:03] Jiao Kang_web_133 joins the room
[11:50:03] Nicolas Kuhn_web_416 joins the room
[11:50:03] Bingyang Liu_web_474 joins the room
[11:50:03] Julien Maisonneuve_web_505 joins the room
[11:50:03] Rui Miao_web_448 joins the room
[11:50:03] Paul Hoffman_web_786 joins the room
[11:50:08] Bingyang Liu_web_474 leaves the room
[11:50:19] Nicolas Kuhn_web_416 leaves the room
[11:51:01] David Millman_web_677 joins the room
[11:51:43] Ayush Mishra_web_490 joins the room
[11:52:21] Nicolas Kuhn_web_543 joins the room
[11:52:34] Dirk Hugo_web_254 joins the room
[11:53:12] Wei Wang_web_621 joins the room
[11:53:35] Spencer Dawkins_web_196 joins the room
[11:53:46] Wei Wang_web_621 leaves the room
[11:53:54] Daniel Havey_web_360 joins the room
[11:54:14] Ingemar Johansson_web_575 joins the room
[11:54:20] J Ignacio Alvarez-Hamelin_web_770 joins the room
[11:54:25] Jana Iyengar_web_257 joins the room
[11:54:37] Markus Amend_web_881 joins the room
[11:54:49] <Ingemar Johansson_web_575> Good afternoon
[11:54:59] Markus Amend_web_881 leaves the room
[11:55:06] Markus Amend_web_186 joins the room
[11:55:10] Theresa Enghardt_web_682 joins the room
[11:55:19] Danyang Chen_web_408 joins the room
[11:55:29] Alessandro Ghedini_web_390 joins the room
[11:55:37] Kazuaki Ueda_web_185 joins the room
[11:55:42] Greg White_web_751 joins the room
[11:56:20] Paul Hoffman_web_786 leaves the room
[11:56:31] <Jana Iyengar_web_257> Hello Everyone!
[11:56:50] <Ayush Mishra_web_490> Good evening from Singapore! :)
[11:57:21] Ike Kunze_web_301 joins the room
[11:57:43] <Rui Miao_web_448> Hello
[11:58:01] Anna Brunstrom_web_767 joins the room
[11:58:06] Roland Bless_web_383 joins the room
[11:58:33] Marten Seemann_web_366 joins the room
[11:58:45] Anthony Faust_web_714 joins the room
[11:58:47] Joakim Misund_web_937 joins the room
[11:59:03] Mitsuaki Hatano_web_123 joins the room
[11:59:04] Tom Jones_web_560 joins the room
[11:59:17] Hirochika Asai_web_268 joins the room
[11:59:19] Michael Scharf_web_508 joins the room
[11:59:22] Richard Scheffenegger_web_474 joins the room
[11:59:29] Alex Huang Feng_web_891 joins the room
[11:59:35] Chris Lemmons_web_927 joins the room
[11:59:53] Hajime Fujita_web_344 joins the room
[12:00:04] Markku Kojo_web_711 joins the room
[12:00:21] Nick Banks_web_546 joins the room
[12:00:28] Bob Briscoe_web_256 joins the room
[12:00:44] Eckard Bogenfeld_web_142 joins the room
[12:00:52] Dan Romascanu_web_479 joins the room
[12:01:06] <Richard Scheffenegger_web_474> say something
[12:01:10] Zaheduzzaman Sarker_web_162 joins the room
[12:01:11] Jake Holland_web_328 joins the room
[12:01:14] <Richard Scheffenegger_web_474> hi
[12:01:15] <Markus Amend_web_186> yes
[12:01:16] Robin Marx_web_290 joins the room
[12:01:18] <Richard Scheffenegger_web_474> yeah!
[12:01:21] <Dirk Hugo_web_254> audio working
[12:01:23] frodek joins the room
[12:01:24] <Chris Lemmons_web_927> Coming through loud and clear.
[12:01:31] Martin Duke_web_966 joins the room
[12:01:34] Kyle Rose_web_176 joins the room
[12:01:37] Bhavit Shah_web_745 joins the room
[12:01:51] Chi-Jiun Su_web_806 joins the room
[12:01:53] Yahya_web_455 joins the room
[12:01:58] <Richard Scheffenegger_web_474> :thumbsup:
[12:01:59] Craig Taylor_web_179 joins the room
[12:02:02] <J Ignacio Alvarez-Hamelin_web_770> Hi!
[12:02:06] <Richard Scheffenegger_web_474> i guess...
[12:02:22] <Kyle Rose_web_176> Good morning!
[12:02:34] Mirja Kühlewind_web_733 joins the room
[12:02:39] <Richard Scheffenegger_web_474> i can not let you down :)
[12:02:48] Marcus Ihlar_web_599 joins the room
[12:02:49] Nigel Davis_web_184 joins the room
[12:02:59] Nigel Davis_web_184 leaves the room
[12:03:06] Nigel Davis_web_794 joins the room
[12:03:46] <Markus Amend_web_186> yes
[12:03:47] Dan Romascanu_web_479 leaves the room
[12:03:49] <Anna Brunstrom_web_767> yes
[12:04:08] Magnus Westerlund_web_224 joins the room
[12:04:27] Vincent Levigneron_web_150 joins the room
[12:04:38] <Chris Lemmons_web_927> A jabber scribe isn't quite as important, in general, when we're all online but having a scribe means that people with disabilities that prevent them from speaking at a mic, or who simply don't have a functional mic, have an opportunity to have things voiced. I'm happy to voice things at the mic. If you want it voiced, just prefix it with "mic:".
[12:04:46] Vincent Levigneron_web_150 leaves the room
[12:05:17] Rodney Grimes_web_916 joins the room
[12:05:35] Gorry Fairhurst_web_130 joins the room
[12:05:40] Nicolas Kuhn_web_543 leaves the room
[12:05:47] Nicolas Kuhn_web_983 joins the room
[12:05:51] Vidhi Goel_web_860 joins the room
[12:06:13] Yeoncheol Ryoo_web_499 joins the room
[12:06:36] John Border_web_387 joins the room
[12:07:01] Spencer Dawkins_web_196 leaves the room
[12:07:04] Spencer Dawkins_web_415 joins the room
[12:07:05] Spencer Dawkins_web_415 leaves the room
[12:07:15] Spencer Dawkins_web_230 joins the room
[12:07:25] Neal Cardwell_web_502 joins the room
[12:07:29] Oh Seokbeom_web_970 joins the room
[12:07:46] james welch_web_331 joins the room
[12:08:30] Jonathan Morton_web_829 joins the room
[12:08:47] Ali Begen_web_569 joins the room
[12:08:58] Rodney Grimes joins the room
[12:09:25] Brian Trammell_web_718 joins the room
[12:09:44] Philip Eardley_web_633 joins the room
[12:09:53] Frode Kileng_web_604 joins the room
[12:10:06] Frode Kileng_web_604 leaves the room
[12:10:09] Frode Kileng_web_388 joins the room
[12:10:22] Stuart Cheshire_web_755 joins the room
[12:11:34] Simon Leinen_web_491 joins the room
[12:11:52] Kazuho Oku_web_553 joins the room
[12:12:03] Nigel Davis_web_794 leaves the room
[12:12:06] Neal Cardwell_web_502 leaves the room
[12:12:09] Neal Cardwell_web_476 joins the room
[12:12:10] Nigel Davis_web_827 joins the room
[12:12:42] Marten Seemann_web_366 leaves the room
[12:12:45] Marten Seemann_web_843 joins the room
[12:13:19] Stephan Emile_web_444 joins the room
[12:13:53] Christian Hopps_web_891 joins the room
[12:14:33] Peter Heist_web_671 joins the room
[12:15:08] Kazuho Oku_web_553 leaves the room
[12:15:40] <Jana Iyengar_web_257> Please add/remove yourself from the mic queue with +q/-q here.
[12:16:29] <Mirja Kühlewind_web_733> jana there is a queue in meetecho...
[12:18:14] <Ingemar Johansson_web_575> Question : L4S gives a finer grain network load info than classic ECN, is this considered ?
[12:19:53] <J Ignacio Alvarez-Hamelin_web_770> mic: Is this protocol intended to be used on the Internet at large or just inside a datacenter?
[12:20:14] <Gorry Fairhurst_web_130> Jana:  In case you missed this: A hugely useful feature is that you can go to the chat panel and click on the second icon - that tears the panel into a window - now a chair can see BOTH the chat and the Mic queue at the same time:grinning:
[12:20:34] <Jana Iyengar_web_257> Yes, of course. Thanks, Mirja -- I forgot about that.
[12:20:36] <Vidhi Goel_web_860> I think in-band telemetry still has more information than Acc ECN.
[12:21:14] <Bob Briscoe_web_256> Regarding HPCC: It occurs to me that, as well as adding telemetry in-band to solve a problem now in DCs, once we see how the DC sender evolves to use the telemetry, we will know which info is /really/ useful - which will inform wider protocol evolution.
[12:21:23] <Spencer Dawkins_web_230> I think this is more comparable to the in-situ OAM work in IPPM than to ECN - among other things, ECN is end-to-end, and it looks like HPCC++ is only within the cloud
[12:21:48] Piers joins the room
[12:22:02] <Mirja Kühlewind_web_733> sounds similar to xcp (which won't work on the Internet but might well in a datacenter)
[12:22:06] <Richard Scheffenegger_web_474> Agree with Vidhi. AccECN has not way to give that much information.
[12:22:07] <J Ignacio Alvarez-Hamelin_web_770> Yes Spencer, is that I having in mind when I send my question
[12:22:16] <Daniel Havey_web_360> If Rui keeps answering out chat questions he will never get through the presentation.
[12:22:21] Yanmei Liu_web_992 joins the room
[12:22:26] <Jana Iyengar_web_257> Thanks Gorry -- I noticed that! Very useful indeed.
[12:22:27] Paolo Saviano_web_144 leaves the room
[12:22:30] Paolo Saviano_web_155 joins the room
[12:22:54] <Ayush Mishra_web_490> @Bob: provided ISPs want to divulge that important information to the endpoints!
[12:23:52] Ahmed Aldabbagh_web_479 joins the room
[12:23:59] Anthony Faust_web_714 leaves the room
[12:24:17] <J Ignacio Alvarez-Hamelin_web_770> The difference here is this protocol is related only to queues, which is useful for traffic and congestion directly. IOAM is measuring other things in other ways, more complicated that what is presented here.
[12:24:39] <Rodney Grimes> There was a paper presented at the Stanford buffer workshop put on by Nick Koen that showed basically you only need 2 bits of feedback to get to a fairly controlled level, and at 4 bits your ROI approaches 0 rapidly.
[12:24:40] <Ayush Mishra_web_490> [mic] what's the difference between the HPCC (presented at SIGCOMM) and HPCC++, which is being presented here?
[12:24:42] <Spencer Dawkins_web_230> @J - yes, exactly
[12:25:10] Anthony Faust_web_172 joins the room
[12:26:17] <Kyle Rose_web_176> If you're not using a headset, please mute when you're not actually speaking because otherwise others' audio echoes back.
[12:26:46] <Spencer Dawkins_web_230> Much better - thanks, Jana!
[12:26:57] <Kyle Rose_web_176> :thumbsup:
[12:27:33] Yanmei Liu_web_992 leaves the room
[12:28:10] Mirja Kühlewind_web_733 leaves the room
[12:28:20] Nitin Batta_web_950 joins the room
[12:29:06] <Jana Iyengar_web_257> Closing the mic queue after Roland
[12:29:07] <Spencer Dawkins_web_230> it's IPPM working group in IETF
[12:29:36] <Spencer Dawkins_web_230> look for -ioam- (if I remember correctly)
[12:29:37] <Bob Briscoe_web_256> @Vidhi, the reason HPCC++ gives diminishing returns for Flow Completion time is 'cos the large part of the inefficiency today is getting up to speed, not running at speed. So for longer flows, the wated time at the start becomes small compared to the total flow completion time.
[12:29:38] <Rodney Grimes> http://buffer-workshop.stanford.edu/papers/paper34.pdf figure 5  shows the deminishing return as the number of bits of feedback increase.
[12:30:26] Yanmei Liu_web_181 joins the room
[12:30:48] <Bob Briscoe_web_256> s/wated/wasted/
[12:31:55] Christian Hopps_web_891 leaves the room
[12:32:09] Christian Hopps_web_466 joins the room
[12:32:16] Eckard Bogenfeld_web_142 leaves the room
[12:33:07] Eckard Bogenfeld_web_174 joins the room
[12:33:11] Jiao Kang_web_133 leaves the room
[12:33:14] Martin Duke_web_966 leaves the room
[12:33:18] Martin Duke_web_420 joins the room
[12:33:29] Jiao Kang_web_313 joins the room
[12:33:30] <Ayush Mishra_web_490> Got it. Thanks, Chris and Rui! It would be interesting to see what practical changes were required to make HPCC actually work.
[12:33:38] Martin Duke_web_420 leaves the room
[12:33:41] Martin Duke_web_169 joins the room
[12:33:58] <Chris Lemmons_web_927> yw
[12:34:06] <Vidhi Goel_web_860> Thanks Bob for the insight. Thanks Stuart for the clarification - I was exactly confused by that. Should have paid attention to the Y-axis.
[12:34:35] Christian Hopps_web_466 leaves the room
[12:34:38] <Bob Briscoe_web_256> @Rod, and Matt Mathis pointed out that, you always use a round trip of feedback, and the faster you go, the more packets per round trip, so the more info you have per round trip, even if you only have a few bits per packet.
[12:34:46] Christian Hopps_web_273 joins the room
[12:35:13] <J Ignacio Alvarez-Hamelin_web_770> @spencer: yes. it is ioam, there are a lot of draft considering different  aspects of the protocol.
[12:38:01] Marten Seemann_web_843 leaves the room
[12:38:04] Marten Seemann_web_232 joins the room
[12:40:30] J Ignacio Alvarez-Hamelin_web_770 leaves the room
[12:40:36] J Ignacio Alvarez-Hamelin_web_367 joins the room
[12:40:44] Nigel Davis_web_827 leaves the room
[12:40:50] <Ingemar Johansson_web_575> On the wish list. #CE marked packets/bytes
[12:41:51] Roland Bless_web_383 leaves the room
[12:41:55] Richard Scheffenegger_web_474 leaves the room
[12:41:56] Roland Bless_web_402 joins the room
[12:42:01] Richard Scheffenegger_web_782 joins the room
[12:42:28] Yahya_web_455 leaves the room
[12:44:28] Nitin Batta_web_950 leaves the room
[12:44:31] Nitin Batta_web_516 joins the room
[12:46:00] <Ingemar Johansson_web_575> Can be useful for e.g L4S deployment testing
[12:46:56] <Rodney Grimes> @Bob, yes exactly, more of that ack thinning can be bad in some situations, as that causes loss of this fairly high bandwidth low bit count feedback mechanism
[12:47:03] Qin Wu_web_137 joins the room
[12:47:36] Christian Hopps_web_273 leaves the room
[12:47:49] Christian Hopps_web_701 joins the room
[12:47:51] <J Ignacio Alvarez-Hamelin_web_367> +q: you could use it form both sides: server or/and client?
[12:48:00] Ian Swett_web_803 joins the room
[12:48:08] <Chris Lemmons_web_927> Ignacio: Do you want that at the mic?
[12:48:09] <Jana Iyengar_web_257> @Ignacio -- Yes
[12:48:34] <Jana Iyengar_web_257> You can in fact have traces on both sides, and the vis tool will merge them for you as well
[12:48:47] <J Ignacio Alvarez-Hamelin_web_367> @chris: it is o.k., just a simple and quick question
[12:48:53] <Chris Lemmons_web_927> ack
[12:49:00] <J Ignacio Alvarez-Hamelin_web_367> thanks
[12:50:04] Nitin Batta_web_516 leaves the room
[12:50:11] Nitin Batta_web_679 joins the room
[12:51:02] <Ayush Mishra_web_490> linux has tcprobe for tracking the cwnd
[12:51:06] <Tom Jones_web_560> I was wondering yesterday during Robins two other talks if we could take tcp blackbox on freebsd and convert to tcp
[12:51:06] <Ayush Mishra_web_490> for TCP
[12:51:52] <Tom Jones_web_560> convert to qlog that is
[12:52:40] <Theresa Enghardt_web_682> I was wondering the same
[12:52:44] <Piers> I thought he mentioned that kprobes could be used to generate TCP info for qlog
[12:53:11] <Tom Jones_web_560> info can come from bpf or dtrace probably quite straght forwardly, but it might have a lot of overhead
[12:53:16] <Stuart Cheshire_web_755> The addition of the ruler function make me laugh! I can’t tell you how many times I’ve held up a straight edge to my screen using tcptrace to work out the slope of the overall transfer!
[12:53:25] <Jake Holland_web_328> yeah, that would be nice.  i think we need better tooling now that more people are digging harder on congestion controls, it doesn't really scale that well for everyone to do special things, it's a pretty heavy lift to do it right.
[12:53:30] <Gorry Fairhurst_web_130> I think the IRTF/IETF should look at this --- I can see lots of potential interest, especially since this can work with DTLS and other layerin.
[12:53:33] <Gorry Fairhurst_web_130> g.
[12:53:51] Alexander Azimov_web_105 joins the room
[12:53:52] Mitsuaki Hatano_web_123 leaves the room
[12:53:52] <Robin Marx_web_290> (I have to admit that I don't know what TCP blackbox is...)
[12:53:58] Mitsuaki Hatano_web_644 joins the room
[12:54:37] <Jonathan Morton_web_829> Department of Redundancy Department
[12:55:38] <Theresa Enghardt_web_682> @Robin: It's event-based logging in the TCP stack in FreeBSD, produces a pcapng file with additional information
[12:56:04] <Tom Jones_web_560> the kernel part is open, sadly the tools havne't yet escaped netflix
[12:56:55] <Robin Marx_web_290> from a quick search, that does look like it might fit qlog/qvis quite well indeed
[12:57:15] <Tom Jones_web_560> is the qlog quic slack a reasonable place to talk about these things?
[12:57:34] <Robin Marx_web_290> yes!
[12:59:56] <Jake Holland_web_328> @robin, have you given any thought to hooking up client-side/server-side traces with a nat?  i assume this is all just connection id on quic, but in tcp i assume the quad woudn't trivially match up from just a pcap import, tho maybe there's concepts like matching flow patterns that could make good guesses...
[13:00:02] <Jana Iyengar_web_257> To the high level q on scaling -- I don't expect qlog itself to be limiting. You can choose a native format for emitting at line rate, and then use offline conversion to qlog when interop with other tooling (such as qvis) is needed. To me, the value of qlog is in separating and producing a standard scheme -- there can be independent evolution on both sides of this schema.
[13:00:18] Joakim Misund_web_937 leaves the room
[13:00:41] Joakim Misund_web_143 joins the room
[13:01:03] <Richard Scheffenegger_web_782> @Tom, Theresa: I would very much like tihs.
[13:01:16] Spencer Dawkins_web_230 leaves the room
[13:01:16] Paolo Saviano_web_155 leaves the room
[13:01:21] Paolo Saviano_web_993 joins the room
[13:01:22] Spencer Dawkins_web_671 joins the room
[13:01:45] Michael Scharf_web_508 leaves the room
[13:01:48] Michael Scharf_web_799 joins the room
[13:02:17] Yoshifumi Nishida_web_141 joins the room
[13:02:21] <Richard Scheffenegger_web_782> yes
[13:02:21] <Martin Duke_web_169> we hear you
[13:02:23] <Markus Amend_web_186> yes
[13:02:24] <Jake Holland_web_328> yes, we hear you
[13:02:25] <Chris Lemmons_web_927> We can hear you
[13:02:59] Andre Bondi_web_926 joins the room
[13:03:56] <Chris Lemmons_web_927> Now you cut out.
[13:03:57] <Richard Scheffenegger_web_782> back
[13:04:02] <Chris Lemmons_web_927> And back
[13:04:36] <Kyle Rose_web_176> FWIW, I have not heard Anna cut out at all.
[13:04:54] <Robin Marx_web_290> @Jake: I don't yet have a good solution for this at this time, no... for QUIC it's indeed much simpler with the CID.
[13:06:13] <Jana Iyengar_web_257> @Stuart -- I have filled out many many sheets of paper with numbers exactly for calculating the slope at different points on the curve :-)
[13:06:14] <Robin Marx_web_290> @Jana: I agree that using converters is definitely an option, but we do still need a decent base format to make it easy to plug-and-play for teams that don't want to invest in conversions
[13:06:22] <Martin Duke_web_169> what are the two segments of each bar?
[13:06:41] <Markus Amend_web_186> end-to-end/tunnel CC
[13:07:12] <Markus Amend_web_186> sry, got your question is wrong
[13:07:26] <Markus Amend_web_186> it's the ration between e2e and tunnel latency
[13:07:28] <Kyle Rose_web_176> What does grey represent vs. (or as a percentage of?) color?
[13:07:34] <Martin Duke_web_169> thanks
[13:07:55] <Kyle Rose_web_176> Ok, makes sense
[13:12:10] Christian Hopps_web_701 leaves the room
[13:14:17] Omer Shapira_web_769 joins the room
[13:14:39] <Martin Duke_web_169> What's the intuition why utilization is so poor with low E2E RTT?
[13:14:47] <Spencer Dawkins_web_671> I swear I thought that was a picture of the Texas coastline during a hurricane :grin:
[13:16:22] Christian Hopps_web_383 joins the room
[13:17:26] Kyle Rose_web_176 leaves the room
[13:17:29] Kyle Rose_web_262 joins the room
[13:17:46] Gorry Fairhurst_web_130 leaves the room
[13:18:07] <Magnus Westerlund_web_224> Jana can you mute Anna?
[13:18:28] <Anna Brunstrom_web_767> In my slide 11, 100% meant full aggregation. was that your question @kyle?
[13:18:32] Gorry Fairhurst_web_994 joins the room
[13:18:46] <Anna Brunstrom_web_767> sorry, muted now, thought I got kicked out
[13:19:12] <Jana Iyengar_web_257> Sorry, I didn't kick you out Anna, I didn't want to be rude :-)
[13:19:35] Michael Scharf_web_799 leaves the room
[13:19:38] Michael Scharf_web_520 joins the room
[13:19:41] Kyle Rose_web_262 leaves the room
[13:19:45] <Anna Brunstrom_web_767> thanks jana :) not used to the this interface
[13:19:47] Kyle Rose_web_413 joins the room
[13:20:09] <Magnus Westerlund_web_224> No, but please mute you are not talking. There wasn't an echo problem, but a typing sound one.
[13:20:40] Michael Scharf_web_520 leaves the room
[13:20:46] Michael Scharf_web_473 joins the room
[13:21:19] Danyang Chen_web_408 leaves the room
[13:21:19] Frode Kileng_web_388 leaves the room
[13:21:20] Jonathan Morton_web_829 leaves the room
[13:21:20] Yeoncheol Ryoo_web_499 leaves the room
[13:21:25] Frode Kileng_web_668 joins the room
[13:21:25] Jonathan Morton_web_520 joins the room
[13:21:26] Yeoncheol Ryoo_web_834 joins the room
[13:21:26] Danyang Chen_web_946 joins the room
[13:21:29] <Jana Iyengar_web_257> @Robin -- you have two contending end goals here: ease of use and efficiency for logging at high speeds. The latter tends to prefer efficient binary data structures in memory that can periodically be batched and emitted, which tends to be at odds with readability and evolvability. All that said, I agree with you that it makes sense to spend some energy on making the schema tighter and more efficient.
[13:21:44] <Anna Brunstrom_web_767> @martin, the reason is that when the RTTs are similar both BBR CCs make similar estimates of the available capacity. so the E2E CC does not increase the send rate enough over the tunnel cc to use the second path. was this your question?
[13:22:05] Hajime Fujita_web_344 leaves the room
[13:22:07] Ian Swett_web_803 leaves the room
[13:22:07] Bob Briscoe_web_256 leaves the room
[13:22:08] Julien Maisonneuve_web_505 leaves the room
[13:22:12] Bob Briscoe_web_274 joins the room
[13:22:14] Julien Maisonneuve_web_643 joins the room
[13:22:14] Ian Swett_web_567 joins the room
[13:22:24] Kyle Rose_web_413 leaves the room
[13:22:36] Hajime Fujita_web_774 joins the room
[13:23:09] <Martin Duke_web_169> yes, thanks
[13:23:19] <Robin Marx_web_290> Christian Huitema has also mused on these aspects for MPQUIC recently in a blogpost, discussing separate packet number spaces: https://huitema.wordpress.com/2021/02/14/how-many-packet-number-spaces-for-quic-multipath
[13:23:49] Kyle Rose_web_309 joins the room
[13:23:53] Brian Trammell_web_718 leaves the room
[13:24:40] <Robin Marx_web_290> @Jana: fully agree
[13:24:54] Yahya_web_201 joins the room
[13:27:41] Piers joins the room
[13:28:05] Piers leaves the room
[13:28:07] <Jana Iyengar_web_257> closing the mic q
[13:29:30] J Ignacio Alvarez-Hamelin_web_367 leaves the room
[13:29:33] J Ignacio Alvarez-Hamelin_web_764 joins the room
[13:30:29] <Theresa Enghardt_web_682> Yes
[13:30:30] <Chris Lemmons_web_927> Yes.
[13:30:32] <Spencer Dawkins_web_671> Markus is remembering SIGTRAN. The question was about "before SIGTRAN" :-)
[13:30:32] <Chris Lemmons_web_927> We can hear.
[13:31:23] Jiao Kang_web_313 leaves the room
[13:31:31] Jiao Kang_web_167 joins the room
[13:32:08] Qin Wu_web_137 leaves the room
[13:32:08] Mirja Kühlewind_web_797 joins the room
[13:32:52] Kiran Makhijani_web_617 joins the room
[13:33:24] J Ignacio Alvarez-Hamelin_web_764 leaves the room
[13:34:40] Michael Scharf_web_473 leaves the room
[13:35:11] Michael Scharf_web_886 joins the room
[13:36:44] <Markus Amend_web_186> A BBR Linux implementation for DCCP, maintained as CCID5. can be found under https://github.com/telekom/mp-dccp/tree/master/net/dccp/ccids
[13:37:58] Kiran Makhijani_web_617 leaves the room
[13:39:03] Piers leaves the room
[13:39:09] Piers joins the room
[13:39:16] Hajime Fujita_web_774 leaves the room
[13:40:16] Hajime Fujita_web_893 joins the room
[13:40:37] Roland Bless_web_402 leaves the room
[13:40:41] <Vidhi Goel_web_860> what's the difference between Swift and BBR.Swift?
[13:40:43] Roland Bless_web_568 joins the room
[13:41:55] <Jana Iyengar_web_257> @Vidhi -- Neal's pres at ICCRG in IETF 109 covers it in more detail
[13:42:05] <Vidhi Goel_web_860> thanks
[13:42:09] Ayush Mishra_web_490 leaves the room
[13:42:15] Ayush Mishra_web_556 joins the room
[13:42:46] <Anna Brunstrom_web_767> could not unmute for some reason
[13:42:47] Paul Congdon_web_888 joins the room
[13:42:54] <Piers> What’s the 2% of Google internal traffic not using BBRv2 ?
[13:43:06] <Chris Lemmons_web_927> Thank you for the presentation. Very informative and interesting.
[13:43:18] <Ingemar Johansson_web_575> +1
[13:43:42] <Neal Cardwell_web_476> The 2% of Google internal traffic not using BBRv2 is a a mix of BBRv1 and the previous datacenter-ECN-based algorithm.
[13:43:53] <Neal Cardwell_web_476> We are working on replacing both of those with BBRv2.
[13:44:09] <Piers> Ok - Thanks
[13:44:56] <Neal Cardwell_web_476> The "datacenter-ECN-based algorithm" I mentioned is the "GCN" mentioned in the Swift SIGCOMM 2020 paper.
[13:45:25] <Anna Brunstrom_web_767> @neal, you mentioned you were still working on getting bbrv2 tuned for the general internet, are there some particular networks scenarios that cause problems atm?
[13:45:39] <Vidhi Goel_web_860> @Neal, if you can share, what are some of the challenges with BBRv2 that you have been recently working on?
[13:46:02] Stephan Emile_web_444 leaves the room
[13:46:11] Stephan Emile_web_389 joins the room
[13:46:33] <Neal Cardwell_web_476> I think the main issues we are running into are related to cases where BBRv2 encounters packet loss. I think the BBRv2 mechanisms for handling packet loss are currently probably too conservative for the public Internet, and need to be tuned.
[13:46:53] <Anna Brunstrom_web_767> thanks
[13:46:53] <Neal Cardwell_web_476> Here "running into" means in the public Internet case for YouTube and google.com
[13:47:27] <Neal Cardwell_web_476> Sorry, I have to sign off now. Thanks, everyone!
[13:47:39] Neal Cardwell_web_476 leaves the room
[13:47:40] <Craig Taylor_web_179> Thank you Neal
[13:47:45] <Bob Briscoe_web_274> Announcement: We've just posted a new first draft writing up the Prague Congestion Control. draft-briscoe-iccrg-prague-congestion-control-00
[13:48:01] Geng-Da Tsai_web_943 joins the room
[13:48:18] <Vidhi Goel_web_860> :thumbsup:
[13:48:27] <Robin Marx_web_290> Clarification question: this assumes a successful address validation when resuming (e.g., using NEW_TOKEN), right?
[13:48:32] Geng-Da Tsai_web_943 leaves the room
[13:50:48] <Marten Seemann_web_232> Clarification question: Why do you need a QUIC extension for this? Why not just encode it into the session ticket?
[13:51:00] Shu-Fang Hsu_web_526 joins the room
[13:51:06] Jiao Kang_web_167 leaves the room
[13:51:08] Kyle Rose_web_309 leaves the room
[13:51:11] Kyle Rose_web_528 joins the room
[13:51:12] Jiao Kang_web_655 joins the room
[13:52:17] <Jana Iyengar_web_257> @Robin: I don't want to answer the question for Gorry, but it has to. It would be considered unsafe to simply use previous CC information for new destination addresses.
[13:53:12] <Robin Marx_web_290> @Jana: yeah, exactly, but that might not be obvious to all readers, so should be stated in the text (which it probably does, haven't looked at it).
[13:54:26] Danyang Chen_web_946 leaves the room
[13:54:37] <Jana Iyengar_web_257> @Marten -- You probably want to do this in the token, but yeah, I don't think this needs a new extension. For this meeting however, it's the CC reuse part of this that's relevant.
[13:54:50] Nick Banks_web_546 leaves the room
[13:56:10] <Robin Marx_web_290> Just checked at https://tools.ietf.org/html/draft-kuhn-quic-0rtt-bdp-08 and it doesn't mention NEW_TOKEN or address validation or amplification prevention. It probably should.
[13:56:25] <Jana Iyengar_web_257> +1
[13:56:44] <Tom Jones_web_560> thanks robin I think that is helpful, could you create an issue? https://github.com/NicoKos/QUIC_HIGH_BDP/
[13:57:20] <Robin Marx_web_290> Will do!
[13:58:04] Ike Kunze_web_301 leaves the room
[13:58:04] Shu-Fang Hsu_web_526 leaves the room
[13:58:46] <Ayush Mishra_web_556> [mic] using the BDP to set the first window relies on the fact that we can actually get an accurate estimate of the BDP. Do we know how good the BDP estimates are? Ideally, we would rather send less than the actual BDP in the first RTT than more than the actual BDP - but perceived BDPs tend to be more than the actual BDP because of queuing.
[13:59:27] Stephan Emile_web_389 leaves the room
[13:59:32] Stephan Emile_web_186 joins the room
[13:59:37] Peter Heist_web_671 leaves the room
[13:59:59] <Jana Iyengar_web_257> We probably won't have time for Q/A -- please take questions to the list. This is an important discussion for QUIC implementations that are very likely to be using something like this.
[13:59:59] <Bob Briscoe_web_274> @Neal, I've been thinking about the problem of loss response too. I wondered whether BBR might find the following useful: An extent-based response to loss using a v inexpensive averaging algo I recently worked out that clocks per RTT but updates per ACK. Reasoning:  rapidly distinguishing the odd loss vs. hitting a wall.
[14:00:01] <Chris Lemmons_web_927> ack
[14:00:38] Jonathan Morton_web_520 leaves the room
[14:00:38] Andre Bondi_web_926 leaves the room
[14:00:43] Jonathan Morton_web_612 joins the room
[14:00:44] Andre Bondi_web_339 joins the room
[14:01:25] Daniel Havey_web_360 leaves the room
[14:01:25] Dirk Hugo_web_254 leaves the room
[14:01:25] Greg White_web_751 leaves the room
[14:01:29] Dirk Hugo_web_690 joins the room
[14:01:29] Greg White_web_412 joins the room
[14:01:30] Daniel Havey_web_855 joins the room
[14:01:31] Dmitry Afanasiev_web_576 joins the room
[14:01:35] Ingemar Johansson_web_575 leaves the room
[14:02:23] Tom Jones_web_560 leaves the room
[14:02:24] Richard Scheffenegger_web_782 leaves the room
[14:02:27] Richard Scheffenegger_web_684 joins the room
[14:02:29] Ahmed Aldabbagh_web_479 leaves the room
[14:02:31] Paul Congdon_web_888 leaves the room
[14:02:32] Ian Swett_web_567 leaves the room
[14:02:33] Theresa Enghardt_web_682 leaves the room
[14:02:34] Yoshifumi Nishida_web_141 leaves the room
[14:02:34] Joakim Misund_web_143 leaves the room
[14:02:34] Markus Amend_web_186 leaves the room
[14:02:34] Ali Begen_web_569 leaves the room
[14:02:35] <Ayush Mishra_web_556> Thank you all!
[14:02:37] Nitin Batta_web_679 leaves the room
[14:02:37] Vidhi Goel_web_860 leaves the room
[14:02:38] Dmitry Afanasiev_web_576 leaves the room
[14:02:38] Frode Kileng_web_668 leaves the room
[14:02:38] Chris Lemmons_web_927 leaves the room
[14:02:38] Yanmei Liu_web_181 leaves the room
[14:02:38] Juhamatti Kuusisaari_web_413 leaves the room
[14:02:39] Alessandro Ghedini_web_390 leaves the room
[14:02:39] Kyle Rose_web_528 leaves the room
[14:02:39] Greg White_web_412 leaves the room
[14:02:40] Bhavit Shah_web_745 leaves the room
[14:02:40] Chi-Jiun Su_web_806 leaves the room
[14:02:41] Martin Duke_web_169 leaves the room
[14:02:41] Jiao Kang_web_655 leaves the room
[14:02:41] Julien Maisonneuve_web_643 leaves the room
[14:02:42] Mitsuaki Hatano_web_644 leaves the room
[14:02:43] Nicolas Kuhn_web_983 leaves the room
[14:02:43] Eckard Bogenfeld_web_174 leaves the room
[14:02:45] frodek leaves the room
[14:02:47] Jana Iyengar_web_257 leaves the room
[14:02:48] Roland Bless_web_568 leaves the room
[14:02:51] Yahya_web_201 leaves the room
[14:02:51] Kazuaki Ueda_web_185 leaves the room
[14:02:52] Andre Bondi_web_339 leaves the room
[14:02:52] <Dirk Hugo_web_690> thanks!
[14:02:55] Alexander Azimov_web_105 leaves the room
[14:02:55] Hajime Fujita_web_893 leaves the room
[14:02:57] Jake Holland_web_328 leaves the room
[14:02:57] Anthony Faust_web_172 leaves the room
[14:03:00] Dirk Hugo_web_690 leaves the room
[14:03:00] james welch_web_331 leaves the room
[14:03:00] Richard Scheffenegger_web_684 leaves the room
[14:03:02] Rodney Grimes_web_916 leaves the room
[14:03:03] Ayush Mishra_web_556 leaves the room
[14:03:08] Magnus Westerlund_web_224 leaves the room
[14:03:09] Daniel Havey_web_855 leaves the room
[14:03:12] Zaheduzzaman Sarker_web_162 leaves the room
[14:03:15] Alex Huang Feng_web_891 leaves the room
[14:03:15] Christian Hopps_web_383 leaves the room
[14:03:19] Michael Scharf_web_886 leaves the room
[14:03:34] Jonathan Morton_web_612 leaves the room
[14:03:41] David Millman_web_677 leaves the room
[14:03:42] Mirja Kühlewind_web_797 leaves the room
[14:03:43] Anna Brunstrom_web_767 leaves the room
[14:03:48] Yeoncheol Ryoo_web_834 leaves the room
[14:04:28] Rodney Grimes leaves the room: Disconnected: BOSH client silent for over 60 seconds
[14:04:44] Markku Kojo_web_711 leaves the room
[14:04:47] Rui Miao_web_448 leaves the room
[14:04:47] Hirochika Asai_web_268 leaves the room
[14:04:47] Robin Marx_web_290 leaves the room
[14:04:47] Craig Taylor_web_179 leaves the room
[14:04:47] Marcus Ihlar_web_599 leaves the room
[14:04:47] John Border_web_387 leaves the room
[14:04:47] Oh Seokbeom_web_970 leaves the room
[14:04:47] Philip Eardley_web_633 leaves the room
[14:04:47] Stuart Cheshire_web_755 leaves the room
[14:04:47] Simon Leinen_web_491 leaves the room
[14:04:47] Marten Seemann_web_232 leaves the room
[14:04:47] Paolo Saviano_web_993 leaves the room
[14:04:47] Spencer Dawkins_web_671 leaves the room
[14:04:47] Gorry Fairhurst_web_994 leaves the room
[14:04:47] Bob Briscoe_web_274 leaves the room
[14:04:47] Stephan Emile_web_186 leaves the room
[14:04:47] Omer Shapira_web_769 leaves the room
[14:17:07] Simon Leinen leaves the room: Stream closed by us: Replaced by new connection (conflict)
[14:17:07] Simon Leinen joins the room
[18:46:25] Piers leaves the room
[19:06:09] Piers joins the room
[19:17:30] Piers leaves the room
[20:24:33] zulipbot joins the room
[20:25:06] zulipbot leaves the room: Disconnected: closed
[20:38:40] zulipbot joins the room
[20:38:57] zulipbot leaves the room: Disconnected: closed
[20:39:00] zulipbot joins the room
[20:39:23] zulipbot leaves the room: Disconnected: closed
[20:39:31] zulipbot joins the room
[20:39:52] zulipbot leaves the room: Disconnected: closed
[20:42:32] zulipbot joins the room
[20:43:36] zulipbot leaves the room: Disconnected: closed
[20:52:04] zulipbot joins the room
Powered by ejabberd - robust, scalable and extensible XMPP server Powered by Erlang Valid XHTML 1.0 Transitional Valid CSS!