Saturday, January 19, 2019

PDS 0.99l release

PDS 0.99l is released with the following changes

1. Fix wrong task priority issue. Now, RT(non-normal) tasks has their proper priority, can be verified by 'RT' priority shown in top/htop.

This is a bug fix release.

Enjoy PDS 0.99l for v4.20 kernel, :)

Code are available at https://gitlab.com/alfredchen/linux-pds
All-in-one patch is available too. 

18 comments:

  1. Builds/boots and works fine. Thanks!

    ReplyDelete
  2. Back from vacation and compiled nohz_full kernel from Oleksandrs git.
    Works fine on i7@work, some tome later will try gaming on Ryzen@home.
    Thanks Alfred and Oleksandr.
    Br, Eduardo

    ReplyDelete
    Replies
    1. nohz_full was no good on Ryzen@home for gaming, frame consistency was bad and it kinda/sorta microstuttered a lot, so I went for nohz_idle w/ 100Hz which was is a lot better. So I'm sticking to nohz_idle at this point.
      Br, Eduardo

      Delete
  3. I saw some previous benchmarks where PDS was comparable with MUQSS+MC. I was wondering if there are more recent benchmarks to help me choose a scheduler (for desktop use, compiling, retrogaming)?

    ReplyDelete
  4. i5 7600k, gigabyte h110m-s2, freezing when i starting dxvk-wine games ( battlefield 4, quake champions )
    in dmesg i'm not have errors

    in journalctl -xb i have only:
    pulseaudio[640]: W: [alsa-source-USB Audio] asyncq.c: q overrun, queuing locally
    pulseaudio[640]: W: [alsa-source-USB Audio] ratelimit.c: 4170 events suppressed
    acpid[426]: client 452[0:0] has disconnected

    and 3 or 4 cores start to load at 100% (in turn)

    on muqss i'm not have problems!(sry for bad english)

    ReplyDelete
    Replies
    1. 100hz, nohz_full

      Delete
    2. Most common suggestion for wine and gaming will be try different yield_type.

      Delete
    3. I use 1000hz, CONFIG_NO_HZ_IDLE. And when gaming i use yield_type=2, rr_interval=1

      I have huge issues with running either MuQSS or PDS with CONFIG_NO_HZ_FULL for some reason, as i think things seem to still be dependent on kernel ticks (perhaps wine more than anything?).

      If you are interested Mr. Anonymous, you can take a look at a example config i made on my github kernel repo for pds and 4.20 kernel. https://github.com/SveSop/kernel_cybmod and look at patch 0014_config.patch.

      Delete
  5. @Alfred Chen, @Sveinar Søpler, thanks guys! yield_type=2, rr_interval=1 helped me!

    ReplyDelete
    Replies
    1. i'm tested bf4 dxvk 0.96, wine-tkg-4.0, + esync + virtual desktop
      in singleplayer
      muqss(rqshare=smt) 90-102fps ( all cores load 100% )
      pds-mq 83-98 ( all cores load max 75-90~ )

      Delete
    2. Please give more details: NOHZ_FULL or NOHZ_IDLE?
      Br, Eduardo

      Delete
  6. Replies
    1. Please try NOHZ_IDLE, from my experience latest PDS / kernel version was not particulary good at gaming, I was getting microstutters a lot and fps was not good w/ NOHZ_FULL...
      At least it might help Alfred to figure out the circumstances of this behaviour and I'll see whethwr this is universally true at this version of PDS.
      Br, Eduardo

      Delete
    2. It is somewhat confusing that ppl are just named "Anonymous", cos it sometimes looks like there is just one person speaking... to himself :)

      I did some testing with the -ck patchset, and ref Con Kolivas the -ck patchset should be configured with NO_HZ_IDLE:
      ---
      I mean intrinsically it doesn't depend on ticks as such, but the configuration of ticks should be nohz idle as you correctly figured out.
      ---
      I do however see that Arch -ck kernel from aur is configured with NO_HZ_FULL. This does not immediately mean it is "the right thing to do".

      Testing PDS vs MuQSS on 4.19 kernel, it seemed as PDS > MuQSS in performance. So far from my testing with 4.20 MuQSS (-ck) is slightly ahead of PDS when it comes to wine gaming atleast. Very minor differences found from the little testing i have done tho, so i guess its "in the eyes of the beholder" :)

      What i DID conclude was atleast on my hardware i had problems with NO_HZ_FULL.

      Delete
  7. I got an error with 4.20.8-rc1

    Setup is 17180 bytes (padded to 17408 bytes).
    System is 5629 kB
    CRC ea1d1ea
    Kernel: arch/x86/boot/bzImage is ready (#1)
    ERROR: "sched_smt_present" [arch/x86/kvm/kvm-intel.ko] undefined!
    make[1]: *** [scripts/Makefile.modpost:92: __modpost]

    https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable-rc.git/commit/?h=linux-4.20.y&id=f4a0d4d3c1532bfa0882f7880764e56c851215fa

    Best regards

    ReplyDelete
    Replies
    1. PDS 0.99m has been released to pick up this sync-up.

      Delete