Announcement

Collapse
No announcement yet.

Dynamic Preemption Support Sent In For The Linux 5.12 Kernel

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

  • Dynamic Preemption Support Sent In For The Linux 5.12 Kernel

    Phoronix: Dynamic Preemption Support Sent In For The Linux 5.12 Kernel

    Ingo Molnar sent in the scheduler updates for Linux 5.12 today and it includes some notable additions, including PREEMPT_DYNAMIC, which allows changing the kernel's preemption mode at boot/run-time...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Does anybody know if tick rate handling of current Windows 10 versions equals Linux' full tickless setting?

    Comment


    • #3
      Originally posted by aufkrawall View Post
      Does anybody know if tick rate handling of current Windows 10 versions equals Linux' full tickless setting?
      How many ticks are you willing to handle and why?

      Comment


      • #4
        Originally posted by aufkrawall View Post
        Does anybody know if tick rate handling of current Windows 10 versions equals Linux' full tickless setting?
        Are you trying to escape Linux?

        I've heard that the NT kernel is set at a 256Hz or 64Hz rate, and I think there is timer coalescing as well...

        It's both the most accurate and most inaccurate kernel for precise timings though.
        If you just sleep and time, it produces millisecond jitter (Linux produces ~100μs jitter).
        If you use HPET or some other hardware interrupt you get almost no jitter.
        Last edited by tildearrow; 17 February 2021, 01:45 PM.

        Comment


        • #5
          Originally posted by tildearrow View Post
          Are you trying to escape Linux?
          Tried that a year ago, didn't work out. It's also pointless to escape Linux with AMD GPU since some games can be played better than on Windows (and some other stuff works better too).

          Comment


          • #6
            That is fantastic new for distros and users. I trust Ingo and other kernel devs that indeed the performance impact compared to a fixed preemption model is close to zero, which make sense with the static patching framework recently introduced.

            Comment


            • #7
              Fantastic news indeed!
              Hopefully Ubuntu will make use of this, since the »lowlatency« kernel flavor is only availabe for x86_64 right now.

              Then again, Ubuntu's standard »generic« flavor uses a 250 Hz timer tick, whereas »lowlatency« relies on 1000 Hz (like Android's kernels).
              Hopefully a way to configure the kernel timer tick at boot-time can also be devised...

              This would basically allow to configure the Linux kernel as a so-called "soft-realtime" one on pretty much any distribution going forward, and not be restricted to just Ubuntu's »lowlatency« kernel flavor, since right now, Canonical is pretty much the only distribution vendor that cares about this (thanks to Ubuntu Studio).

              Comment


              • #8
                Originally posted by Linuxxx View Post
                Hopefully a way to configure the kernel timer tick at boot-time can also be devised...
                Or even a syscall that allows entering an (adaptive) tick mode for a specific CPU or the calling thread. If that's what you meant. (Or a writeable sysfs flag, if writing to sysfs doesn't interfere with any mode.)

                If you meant the tick frequency, why not an administrator command that can change the frequency?
                Last edited by indepe; 17 February 2021, 05:37 PM.

                Comment


                • #9
                  This is good news! I have a custom kernel build on a pc just because that pc runs KVM with GPU passthrough. And in that situation you want to change the PREEMPT to be voluntary because otherwise your boot (of a virtual windows 10) can take super long! Depending on how much ram you assign to it.

                  The downside is that i have a custom kernel that i need to rebuild every time the host os has a kernel update I prefer to stay on the host kernel from the package manager and not go the custom compile route.

                  Comment

                  Working...
                  X