Announcement

Collapse
No announcement yet.

Kolivas Takes Break From Designing COVID-19 Equipment To Release Linux 5.6-ck1 + MuQSS

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Kolivas Takes Break From Designing COVID-19 Equipment To Release Linux 5.6-ck1 + MuQSS

    Phoronix: Kolivas Takes Break From Designing COVID-19 Equipment To Release Linux 5.6-ck1 + MuQSS

    Con Kolivas is out with his Linux 5.6-ck1 optimization patch-set and version 0.199 of the MuQSS scheduler. This re-base against the Linux 5.6 stable kernel is coming late due to Kolivas leading a team making 3D printed COVID-19 equipment in Australia...

    http://www.phoronix.com/scan.php?pag...-5.6-ck1-MuQSS

  • #2
    What a hero :heart:

    Comment


    • #3
      I am out of the loop here: are his patches universally considered beneficial to kernel performance?
      If yes, why aren't they merged to upstream?
      If no, what are the issues/drawbacks compared to mainline kernel?

      Comment


      • #4
        Originally posted by turboNOMAD View Post
        I am out of the loop here: are his patches universally considered beneficial to kernel performance?
        If yes, why aren't they merged to upstream?
        If no, what are the issues/drawbacks compared to mainline kernel?
        I think his patches are mainly with the intention to boost performance of interactive desktop loads where you multitask many applications and want good performance as in low latency in the interaction. It the intended target is not server loads that focus more on high throughput.

        Comment


        • #5
          I am using this kernel in Arch Linux and indeed noticed responsiveness improvements. That said, it seems that 5.6 kernels are also better in that regards than older kernels.
          You may also try: zen and liquorix that also try to optimise desktop usage. Unfortunately there is very few means to benchmark them other than reporting a "feeling".

          Comment


          • #6
            On 5.5-ck1 on an FX-8350 I have experienced significantly slower start up times while compiling (even with nice -n19 and ionice -c3 on the compilation load) and trying to open a konsole when using ck1, muqss and the bfq low latency I/O scheduler together. I also suffer from lower-pitch audio when playing videos.

            It is my understanding that -ck1 is not well tested on AMD hardware and additionally, I'm certainly not ruling out a local (mis)configuration issue on my part.

            Comment


            • #7
              Originally posted by turboNOMAD View Post
              If yes, why aren't they merged to upstream?
              If no, what are the issues/drawbacks compared to mainline kernel?
              It doesn't play with some software when CPU cgroups get involved. Some python ionice tool, docker, earlyOOM(might have been oomd), schedutil, BFQ(not sure about this one, I know there is some sort of cgroup stuff, but it might not be the kind that are affected), etc. Some don't need cgroup support, but if they detect it in the kernel they'll try to make use of it, but MuQSS just stubs them, as does BMQ. I think another one affected by the lack of this particular cgroup support was a dedupe tool for BTRFS.

              So uhh.. stuff your distro may provide out of the box these days, or you may use for work or as a power user can potentially break. Whatever OOM project it was, I remember someone reporting it going out of whack and it was because of the metrics were no longer reliable(something in the kernel config messes with CPU accounting, and also throws off monitoring for accurate CPU activity % iirc), so the software wasn't working as intended. At one point at least it also broke docker, so if you used that for development or hobby, it was probably more important to you than the improvements MuQSS would offer.

              Neither MuQSS or BMQ devs have time to sort that out for years afaik, so it's not likely to be addressed.

              Comment

              Working...
              X