Announcement

Collapse
No announcement yet.

BFS Scheduler Update Brings SMT Nice Support

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

  • BFS Scheduler Update Brings SMT Nice Support

    Phoronix: BFS Scheduler Update Brings SMT Nice Support

    Con Kolivas released a new version of his BFS scheduler and besides porting it for Linux 3.16 compatibility it also contains a big new feature...

    http://www.phoronix.com/vr.php?view=MTc2NDQ

  • #2
    Why isn't this in the main kernel? By this I mean BFS and BFQ

    Comment


    • #3
      Originally posted by Vidar View Post
      Why isn't this in the main kernel? By this I mean BFS and BFQ
      A)CFS works better for CPU batch tasks, e.g. servers
      B)There has been a dispute between Con Kolivas and the other Linux developers (which caused his resigning)

      Comment


      • #4
        Originally posted by Nognir View Post
        A)CFS works better for CPU batch tasks, e.g. servers
        But isn't BFS better for desktop usage? If so it seems logical to allow users to choose different schedulers depending on their computer and needs.

        Comment


        • #5
          Originally posted by Vidar View Post
          Why isn't this in the main kernel? By this I mean BFS and BFQ
          It was planned to merge BFQ in the main kernel, but there haven't been any news about that since June.

          Comment


          • #6
            Originally posted by Vidar View Post
            But isn't BFS better for desktop usage? If so it seems logical to allow users to choose different schedulers depending on their computer and needs.
            Certain people have claimed that, but i don't think anyone has been able to show it definitively. The people in charge of the scheduler code certainly don't believe so, and it doesn't look like most of the distro's kernel maintainers do either, although you can find a few lesser known distros that use it.

            Comment


            • #7
              Originally posted by Vidar View Post
              But isn't BFS better for desktop usage? If so it seems logical to allow users to choose different schedulers depending on their computer and needs.
              This is unreliable and unsupported but the last I remember reading, BFS was only good for 4 cores or less. After 4 cores the CFS scheduler wins everything.

              Since even my phone has 4 cores and my desktop has 6, and the next desktop I buy will probably have 8 cores, I don't think BFS is a good future investment.

              Comment


              • #8
                TBH, I tried compiling the linux-ck kernel from the AUR (Arch user here) and there was hardly any noticable difference. And I'm running a 4 year old Lenovo AMD dual core laptop

                Comment


                • #9
                  Originally posted by Vidar View Post
                  Why isn't this in the main kernel? By this I mean BFS and BFQ
                  Quotes from ck FAQ (http://ck.kolivas.org/patches/bfs/bfs-faq.txt)
                  Are you looking at getting this into mainline?

                  LOL.

                  No really, are you?

                  LOL.

                  Really really, are you?

                  No. They would be crazy to use this scheduler anyway since it won't scale to
                  their 4096 cpu machines. The only way is to rewrite it to work that way, or
                  to have more than one scheduler in the kernel. I don't want to do the former,
                  and mainline doesn't want to do the latter. Besides, apparently I'm a bad
                  maintainer, which makes sense since for some reason I seem to want to have
                  a career, a life, raise a family with kids and have hobbies, all of which
                  have nothing to do with linux.

                  Comment


                  • #10
                    Originally posted by Zan Lynx View Post
                    This is unreliable and unsupported but the last I remember reading, BFS was only good for 4 cores or less. After 4 cores the CFS scheduler wins everything.

                    Since even my phone has 4 cores and my desktop has 6, and the next desktop I buy will probably have 8 cores, I don't think BFS is a good future investment.
                    Yep, just looked at the wikipedia article about this again and a linked benchmarking (3.6.2 kernel) and up to i7-3770k, i.e. 1-4 cores showed small gains, the xeon 12 core (I think that this was 12 logical -> 6 physical) showed quite a significant drop, ~1.7% v. CFS while the others IIRC were c. 0.5- ~1% increase... big whoop...

                    Granted the benching was pretty limited in what was tested, but...

                    I really suppose the best way to do it, is just to try it out and see which scheduler works best for whatever it is that you do, but with those tiny differences it probably won't be a "feel"(subjective) evaluation, unless there are some cases not covered where yes indeed BFS/Q are better for typical/dev desktop usage while CFS/whatever is better for XYZ server(of course I expect that will highly depend upon what the server is actually being used for) IOW there is probably no cut and dried answer.

                    To me, though, the tiny differences aren't worth my time experimenting with. Maybe when I do a haswell-e build I'll experiment on that before making it my primary desktop, or the demoted 3930k... (which will probably become a <something>unixish(might be *BSD instead of linux as I still prefer *BSD for servers, and I won't care about sound/graphics/video/etc. drivers/support as it'll be headless/speakerless/kbless/etc.less) server.

                    Comment

                    Working...
                    X