[{"author": "Reese Enghardt", "text": "<p>@Meetecho Camera on speaker, please</p>", "time": "2026-07-22T07:06:33.000Z"}, {"author": "Alan Jowett", "text": "<p>244us is approximately the reciprocal of 4096</p>", "time": "2026-07-22T07:11:09.000Z"}, {"author": "Alan Jowett", "text": "<p>so, 4096 times per second == 244us</p>", "time": "2026-07-22T07:11:25.000Z"}, {"author": "Michael Welzl", "text": "<p>Yes - probably:</p>\n<p>1) less than 1ms, to avoid the bursts for L4S as Christian said<br>\n2) why not 1/4, for example;<br>\n3) let's implement 1/4 efficiently in the kernel.</p>\n<p>That makes sense to me.</p>", "time": "2026-07-22T07:17:45.000Z"}, {"author": "Christian Huitema", "text": "<p>The I in PIE is for integral, ie., assessing the likely impact of past feedback on the various flows. That requires guessing both the RTT of the flows and the response mode of these flows. I wonder what happens if these guesses are wrong.</p>", "time": "2026-07-22T07:27:30.000Z"}, {"author": "Mohit Tahiliani", "text": "<p>Yes, but the integral component accumulates error over time rather than instantaneously, so minor RTT or response mode errors might not impact largely. There is an auto-tuning feature as well in RFC 8033 (PIE) that automatically scales the controller parameters depending on the drop probability.</p>", "time": "2026-07-22T07:44:16.000Z"}, {"author": "Christian Huitema", "text": "<p>Might not =&gt; maybe measure that?</p>", "time": "2026-07-22T07:46:27.000Z"}, {"author": "Mohit Tahiliani", "text": "<p>Yes, we have tested it in varying RTT scenarios, and with different response modes.</p>", "time": "2026-07-22T08:00:57.000Z"}, {"author": "Christian Huitema", "text": "<p>Did you publish the results?</p>", "time": "2026-07-22T08:01:38.000Z"}, {"author": "Mohit Tahiliani", "text": "<p>Yes, I'll share the documents with you.</p>", "time": "2026-07-22T08:08:12.000Z"}, {"author": "Chris Box", "text": "<p>Which BBR?</p>", "time": "2026-07-22T08:23:56.000Z"}, {"author": "Christian Huitema", "text": "<p>Thanks, Mohit.</p>", "time": "2026-07-22T08:24:10.000Z"}, {"author": "Chris Box", "text": "<p>I'm guessing 2019 means version 1</p>", "time": "2026-07-22T08:24:38.000Z"}, {"author": "Ike Kunze", "text": "<p>@Chris: yes, that paper was about version 1</p>", "time": "2026-07-22T08:27:01.000Z"}, {"author": "Chris Box", "text": "<p>thx</p>", "time": "2026-07-22T08:27:10.000Z"}, {"author": "Michael Welzl", "text": "<p>This (performance isolation) sounds like a system that could give people all we need from L4S, simpler.</p>", "time": "2026-07-22T08:32:52.000Z"}, {"author": "Christian Huitema", "text": "<p>You mean DualQ AQM? Yes, this presentation seems more interesting.</p>", "time": "2026-07-22T08:33:41.000Z"}, {"author": "Christian Huitema", "text": "<p>Also, without using equations hat used to model TCP 30 years ago....</p>", "time": "2026-07-22T08:34:34.000Z"}, {"author": "Michael Welzl", "text": "<p>Yes. The whole instantaneous marking, Prague reaction part of L4S is only relevant for the many (irony alert) delay-sensitive but also queue-building flows that exist...</p>", "time": "2026-07-22T08:34:35.000Z"}, {"author": "Chris Box", "text": "<p>Is \"most of the internet\" using FIFO? Not sure that is true of access network bottlenecks.</p>", "time": "2026-07-22T08:36:48.000Z"}, {"author": "Mohit Tahiliani", "text": "<p>Flow Queuing  separates old and new flows, and also prioritize the ones that are sparse and do not have a standing queue</p>", "time": "2026-07-22T08:36:59.000Z"}, {"author": "Michael Welzl", "text": "<p>Yeah but this proposed system scales better.</p>", "time": "2026-07-22T08:38:24.000Z"}, {"author": "Christian Huitema", "text": "<p>I would be happy with two queues, one for CCA that don't do bufferbloat and one for those that do...</p>", "time": "2026-07-22T08:40:08.000Z"}, {"author": "Michael Welzl", "text": "<p>+1 !</p>", "time": "2026-07-22T08:41:04.000Z"}, {"author": "Michael Welzl", "text": "<p>I mean, I agree, not \"+1\" in: I suggest 3  :-D</p>", "time": "2026-07-22T08:41:29.000Z"}, {"author": "Christian Huitema", "text": "<p>Boo</p>", "time": "2026-07-22T08:42:35.000Z"}, {"author": "Christian Huitema", "text": "<p>(Boo to Jana)</p>", "time": "2026-07-22T08:42:45.000Z"}, {"author": "Christian Huitema", "text": "<p>+1 This is great work!</p>", "time": "2026-07-22T09:00:27.000Z"}, {"author": "Reese Enghardt", "text": "<p>Thanks all!</p>", "time": "2026-07-22T09:01:21.000Z"}]