LWN: Comments on "Deadline scheduler part 2 — details and usage" https://lwn.net/Articles/743946/ This is a special feed containing comments posted to the individual LWN article titled "Deadline scheduler part 2 — details and usage". en-us Tue, 21 Oct 2025 09:00:38 +0000 Tue, 21 Oct 2025 09:00:38 +0000 https://www.rssboard.org/rss-specification lwn@lwn.net Deadline scheduler part 2 — details and usage https://lwn.net/Articles/929189/ https://lwn.net/Articles/929189/ vssasks <div class="FormattedComment"> It's not clear how to calculate deadline parameters for threads of a daw running for example at 44000/16bit. Its parent process creates multiple threads: an audio thread running at fifo 99 (defined manually in preferences), a lot of threads each of which is created for each audio plugin loaded with fifo 1, one midi and one video thread with nonrt prio 20. Since daw is very cpu intensive, the common approach is to run all those threads on all cpus but one left for housekeeping. Will there be benefit to use deadline scheduler for all threads except audio thread which priority in any case will be reset to that set in preferences? It would be interesting to experiment with that if i knew how to calculate parameters for the command. The buffer for audio thread is set at 16/3 which equals to latency of 1 ms<br> </div> Sun, 16 Apr 2023 03:16:04 +0000 Deadline scheduler part 2 — details and usage https://lwn.net/Articles/745788/ https://lwn.net/Articles/745788/ voodoosound <div class="FormattedComment"> I found the problem. For anyone having a hard time with this too.<br> First of all manpages are your friend. Read the error description.<br> <p> I had overlooked the second note on EPERM.<br> It says that all CPUs have to be available to the scheduler ( e.g. sched_setaffinity ). Otherwise it fails with EPERM.<br> <p> Hope this helps.<br> <p> BR,<br> Christoph<br> </div> Tue, 30 Jan 2018 17:46:41 +0000 Shared ressources https://lwn.net/Articles/745745/ https://lwn.net/Articles/745745/ jezz <div class="FormattedComment"> In add to Dhall effect, pitfalls related to critical sections should be mentioned. A task with a close deadline may be blocked with a task with more distant deadline. Unfortunately, shared resources are everywhere in modern OS and tasks inter-blocking latencies are difficult to evaluate (especially with dynamic priority schedulers).<br> </div> Tue, 30 Jan 2018 11:07:53 +0000 Deadline scheduler part 2 — details and usage https://lwn.net/Articles/745626/ https://lwn.net/Articles/745626/ voodoosound <div class="FormattedComment"> Hi,<br> <p> great article! Convinced me to subscribe!<br> <p> I wanted to test this.<br> But setting the SCHED_DEADLINE policy with sched_setattr() fails with an operation not permitted error.<br> SCHED_RR works fine this way.<br> <p> I am running this Kernel:<br> Linux srv2 4.8.6-rt5 #2 SMP PREEMPT RT Wed Nov 23 21:12:52 CET 2016 x86_64 x86_64 x86_64 GNU/Linux<br> <p> Is there anything I have to setup in the system?<br> <p> <p> BR,<br> Christoph<br> </div> Mon, 29 Jan 2018 14:19:03 +0000 Deadline scheduler part 2 — details and usage https://lwn.net/Articles/744952/ https://lwn.net/Articles/744952/ rvfh <div class="FormattedComment"> You lost me on this one. Maybe a graph or two would help?<br> </div> Mon, 22 Jan 2018 13:51:10 +0000 Deadline scheduler part 2 — details and usage https://lwn.net/Articles/744882/ https://lwn.net/Articles/744882/ atelszewski <div class="FormattedComment"> Hi,<br> <p> This is all quite an interesting reading for embedded systems guy like me :-)<br> Well appreciated!<br> <p> I just would like to emphasize more the RT patches, because the people behind it<br> deserve a big credit.<br> <p> It's true what you wrote about RT preventing lower tasks from (sigh) preventing<br> higher priority tasks from running when they need to. But the real cool and hardest<br> thing in all the RT patches is making the kernel as lock-less and as preemptible<br> as possible, including all the drivers. And all the crazy possible code paths you can<br> only imagine ;-)<br> <p> --<br> Best regards,<br> Andrzej Telszewski<br> </div> Sun, 21 Jan 2018 23:55:41 +0000