Wednesday, January 17, 2018

PDS 0.98j release (respinned)

PDS 0.98j is released with the following changes

1. Sync up with mainline 4.15 scheduler code changes. (Updated to sync with official 4.15 mainline)

This is just a sync up release for 4.15 kernel. I will be on a trip and may not able to release new PDS code for kernel 4.15 when it comes out next week. So it is released earlier this week, the sync up changes should be completed if there is no last minute scheduler changes from mainline. Git repositories will be updated later.

Enjoy PDS 0.98j for v4.15 kernel, :)

code are available at
https://bitbucket.org/alfredchen/linux-gc/commits/branch/linux-4.15.y-pds
and also
https://github.com/cchalpha/linux-gc/commits/linux-4.15.y-pds

 All-in-one patch is available too.

18 comments:

  1. Awesome Alfred; I had meant to ask if you were working on a 4.15 sync. I will try this out soon.

    ReplyDelete
  2. Build error: https://gist.github.com/fd3789b5d179c04d525a8435cf7c68b6
    Fix: https://gist.github.com/4339e7b61d221f712ff252740296e1b9

    ReplyDelete
    Replies
    1. Oh, that should be the last minute changes, :)
      I haven't seen it in 4.15-rc8.

      Delete
    2. Yeah, this change was introduced later:

      $ git describe c96f5471ce7d2
      v4.15-rc8-1-gc96f5471ce7d

      Delete
    3. Ha! I just did the same thing & sent mail to Alfred but for 4.14.15, since that change was queued for 4.14-stable as well..

      Delete
    4. Thanks for the 4.14.15 related notification! Need to keep this kernel around for some more time. BR, Manuel Krause

      Delete
    5. @pf
      Thanks for the info and your fix patch. Good news is that 4.15 is delayed. I will check latest mainline code changes next week. Before that, your fix should be good.

      Delete
  3. Finally getting 4.15rc9 going w/ PDS...
    Alfred (or Olek), any comment about the CPU_ISOLATION change, regarding desktop/interactive usage?
    https://github.com/torvalds/linux/commit/5c4991e24c69
    https://patchwork.kernel.org/patch/10085607/

    ReplyDelete
    Replies
    1. Update: x86-UP built fine; haven't booted it yet but I'm running rc9 x64 now:
      Linux version 4.15.0-1-pds (gcc version 7.2.1 20171224 (GCC)) #1 SMP PREEMPT
      Planning to try linux-next once it gets updated again.

      Delete
    2. I may try to build NOHZ_FULL some time, which in my mind should be superior to NOHZ_IDDLE, but so far it was not...
      It would be interesting if isolation helps...

      Delete
  4. Since it's been quiet here...I have run 4.15.2 (with all the latest BFQ patches and some other linux-block stuff) on both x64 (Athlon X2 5200+) and x86 no-SMP (Atom netbook), including suspend/resume, without issue.

    ReplyDelete
    Replies
    1. No news is good news. At least no bug reports to be worried about.
      Recently, I am doing another round of code clean-up and testing some new ideas. Next release is most likely just a code clean-up release, :)

      Delete
    2. Cool; btw I also test a linux-next config on occasion, so if you end up working on 4.16 early, I can help test if you want. :-)

      Delete
    3. @Alfred:
      A side question regarding your next improvements:
      I assume that you've read into and checked code about what Con Kolivas did with "Runqueue sharing experiments with MuQSS." {http://ck-hack.blogspot.de/2017/11/runqueue-sharing-experiments-with-muqss.html}
      Would the advantages be applicable/ useful for your PDS too?

      BR, Manuel Krause

      Delete
    4. @Manuel
      It is an interesting topic. It's a brilliant idea to reconsider how to manage the run queue in the view of physical cpu or llc rather than from the view of the logic cpu(currently). I am not saying that it must be better, but you always find something new by looking thing in different point of view, isn't it?
      In PDS, what I am first think of is the advantages by reducing number of run queues. As current pds already consider the scalability of growing logic cpus, so there may be 2 to 3 spot of code may be benefit by reducing number of run queues. Considering the challenge work to adapt to sharing runqueue, I'd like to put effects on current per cpu runqueue design first, then look back into share runqueue implementation when run out of idea based on current design, :)

      Delete
    5. :-) @ Alfred
      Based on your past work I'm absolutely sure that you'll find the best next steps to improve 'your baby', PDS.
      (ATM, I'm still with the 4.14 PDS kernel -- and no issues at all. Thanks for that!)

      BR, Manuel Krause

      Delete
    6. Manuel, are you on 4.14 due to ToI issues, or other? Fwiw Alfred, I'm on 4.15.3 with PDS-j and latest BFQ patches here, also no issues. Was worried after a couple times last week I saw the suspended PC was off the next day...but then I started getting random power off. Not a shutdown...like the power was switched off. I swapped in my spare repaired PSU and all good again; now to repair the bad one. :-/ I will update to -k tonight.

      Delete
    7. @jwh:
      Yes, you're right. Atm. I'm unable to load a proper TOI patch for 4.15 via the github web interface (and am too lazy to go through the true git steps for a local copy and cleaning free some disk space before). I've notified Nigel about it and maybe sooner or later he finds some time...

      Regarding the PDS 0.98k I can say that the commits from 49d5b8318f59d34eb6dfc946a174c79300024d32 and newer also apply fine on 4.14.19 (only little offsets) on top of 0.98i and work well for now ~22h on here.

      BR, Manuel Krause

      Delete