Announcement

Collapse
No announcement yet.

Torvalds' Comments On Linux Scheduler Woes: "Pure Garbage"

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

  • Torvalds' Comments On Linux Scheduler Woes: "Pure Garbage"

    Phoronix: Torvalds' Comments On Linux Scheduler Woes: "Pure Garbage"

    As you may recall a few days ago there was the information on the Linux kernel scheduler causing issues for Google Stadia game developers. The scheduler was to blame and in particular Linux's spinlocks. Linus Torvalds has now commented on the matter...

    http://www.phoronix.com/scan.php?pag...Scheduler-Woes

  • #2
    It's 2020 and the old, grumpy Torvalds is back. Oh, how I missed his rudeness I have to say that the Linux scheduler definitely needs improvements though. BMQ works so much better in terms of system responsiveness when compiling packages and frametimes in games.
    Last edited by kiffmet; 01-05-2020, 09:09 AM.

    Comment


    • #3
      I do fully agree on the complexity of locking and the stupidity of inventing your own mechanism for that.
      I have however, also seen very, very odd scheduling behavior from the Linux kernel.
      Stuff I cannot explain other than "looks brain dead".

      It's probably easy to just blame all the loads as "stupid and utter garbage", which probably would be true in a lot of cases.
      But. Here comes my but. Since Linux does NOT HAVE ANY historic performance metric proving anything,
      all claims of "performance and fairness" from the kernel side also become moot and garbage claims.

      In fact. A lot of code and patches go into core kernel bits without ANY metric whatsoever.
      Heck even Phoronix has to bisect the kernel all the time for major performance regressions. And that's only the major bits.

      I really, really think the kernel could do with better proving grounds. They would also serve as an excellent implementation examples.
      Last edited by milkylainen; 01-05-2020, 09:14 AM.

      Comment


      • #4
        Originally posted by kiffmet View Post
        It's 2020 and the old, grumpy Torvalds is back. Oh, how I missed his rudeness
        I agree. But I wish his outbursts were more productive.
        Like.
        "You're stupid. But. WTH. Here you go. Now we have implemented a pretty full-featured userspace locking test and performance metrics mechanism with x & y & z."
        "This will be used for validation."

        Comment


        • #5
          Originally posted by kiffmet View Post
          It's 2020 and the old, grumpy Torvalds is back. Oh, how I missed his rudeness
          How is he rude in this case? I actually never thought he was rude, it can be frustrating working with people when you're simply not good at it, so that's why his comments used to be the way they were. And he always confronted the problem, not just the person.

          Comment


          • #6
            Originally posted by kiffmet View Post
            It's 2020 and the old, grumpy Torvalds is back. Oh, how I missed his rudeness I have to say that the Linux scheduler definitely needs improvements though. BMQ works so much better in terms of system responsiveness when compiling packages and frametimes in games.
            Windows is not better here though, also games utilizing only few threads then show heavy stutter as well.

            Comment


            • #7
              ”room for improvement” is #1 on my loathe-list of corporate lingo.

              Comment


              • #8
                Originally posted by phoronix View Post
                Phoronix: Torvalds' Comments On Linux Scheduler Woes: "Pure Garbage"

                As you may recall a few days ago there was the information on the Linux kernel scheduler causing issues for Google Stadia game developers. The scheduler was to blame and in particular Linux's spinlocks.

                http://www.phoronix.com/scan.php?pag...Scheduler-Woes
                @michael
                That's wrong! It's not about spinlocks from Linux. The developer who blamed the kernel has written his own spinlock in userspace without necessary knowledge. He did his own locking wrong and was measuring the wrong numbers. Spinlocks from Linux kernel are completely out of scope here.

                Comment


                • #9
                  Originally posted by milkylainen View Post

                  I agree. But I wish his outbursts were more productive.
                  Like.
                  "You're stupid. But. WTH. Here you go. Now we have implemented a pretty full-featured userspace locking test and performance metrics mechanism with x & y & z."
                  "This will be used for validation."
                  If you read the actual thread you will see he is being more productive in later posts. The point he was trying to make is that what this person was trying was fundamentally flawed.

                  Comment


                  • #10
                    In Linus we trust!

                    Comment

                    Working...
                    X