Announcement

Collapse
No announcement yet.

Sched_ext Merged For Linux 6.12 - Scheduling Policies As BPF Programs

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

  • #11
    Originally posted by gotar View Post
    This is a tool for hyperscalers, just like many other "pro" features in recent years. Have you ever seen any XDP/CPUMAP in the wild, used by some random admin? No? Google was glad having it...
    Petabyte storages, faster booting with hundreds of CPUs, Wayland (closed-systems display, e.g. automotive)... see the pattern?

    These things are often sponsored by Big Corps. And to be clear - it's good we have it for free, despite we probably won't ever use them (directly, not as a service...), as they require (re)building some background (e.g. folios, DRI) and there are the real profits for everyone.
    Not in this case that is just for hyperscalers for this feature..
    sched_ext schedulers and tools. Contribute to sched-ext/scx development by creating an account on GitHub.

    Ubuntu developers working on this some of those developers have been focused on gaming workloads.

    The means to load and unload a custom scheduler has many different uses. Yes I can see in time people getting banned from games because game server thinks they are cheating because they are using a custom scheduler matched to game so their game is running better. Yes and the custom scheduler just happen to drop the priority completely off the anti-cheat software.

    Comment


    • #12
      Originally posted by oiaohm View Post
      Not in this case that is just for hyperscalers for this feature..
      sched_ext schedulers and tools. Contribute to sched-ext/scx development by creating an account on GitHub.

      Ubuntu developers working on this some of those developers have been focused on gaming workloads.

      OK, let me quote from above link: "Both Meta and Google are fully committed to sched_ext and Meta is in the process of mass production deployment".



      So if any gaming or other end-user scheduler emerges for wide and common use, that will be a "collateral profit".

      Comment


      • #13
        Originally posted by gotar View Post
        that will be a "collateral profit".
        Well, lots of desktop Linux enhancements nowadays are a "collateral profit". That's not 2007-2014 anymore, when Linux was blooming with one-man projects.

        Comment


        • #14
          Originally posted by gotar View Post

          OK, let me quote from above link: "Both Meta and Google are fully committed to sched_ext and Meta is in the process of mass production deployment".
          So if any gaming or other end-user scheduler emerges for wide and common use, that will be a "collateral profit".
          I have the distortion towards a perception that this would be most helpful in regularizing and crystallizing the patterns of behavior upon the levels of desktop experience. And so the evolution of the Linux desktop remains rather unpredictable.

          Comment


          • #15
            Originally posted by gotar View Post

            This is a tool for hyperscalers, just like many other "pro" features in recent years. Have you ever seen any XDP/CPUMAP in the wild, used by some random admin? No? Google was glad having it...
            Petabyte storages, faster booting with hundreds of CPUs, Wayland (closed-systems display, e.g. automotive)... see the pattern?

            These things are often sponsored by Big Corps. And to be clear - it's good we have it for free, despite we probably won't ever use them (directly, not as a service...), as they require (re)building some background (e.g. folios, DRI) and there are the real profits for everyone.
            I use XDP at work for extremely low latency filtering.
            This is absolutely awesome, as it allows supported NICs to run the XDP program on the NIC itself.

            Comment


            • #16
              Originally posted by gotar View Post

              This is a tool for hyperscalers, just like many other "pro" features in recent years.
              Nah. This particular feature has big implications for any regular users interested in gaming. So it isn't just about tuning for some hyperscaler workloads.

              Comment


              • #17
                Originally posted by pWe00Iri3e7Z9lHOX2Qx View Post

                Nah. This particular feature has big implications for any regular users interested in gaming. So it isn't just about tuning for some hyperscaler workloads.
                "Has" (in present perfect), or maybe some future conditional? Is there any Valve involvement? Are there already some SUID-games? are schedulers distributed with games, or some gaming platform has one common? Excuse my ignorance, I don't play games.

                Comment


                • #18
                  Originally posted by gotar View Post

                  "Has" (in present perfect), or maybe some future conditional? Is there any Valve involvement? Are there already some SUID-games? are schedulers distributed with games, or some gaming platform has one common? Excuse my ignorance, I don't play games.
                  Fair questions, and your first assumption is why it will likely go somewhere real that millions of people will use.

                  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


                  Especially on weaker devices like the Steam Deck, the improvements can be meaningful.

                  Comment


                  • #19
                    Originally posted by gotar View Post

                    "Has" (in present perfect), or maybe some future conditional? Is there any Valve involvement? Are there already some SUID-games? are schedulers distributed with games, or some gaming platform has one common? Excuse my ignorance, I don't play games.
                    Well Google makes phone software and web browsers.
                    I would think that a CPU scheduler that is working well on a phone device, and good for browsing the web, probably also is good for gaming.

                    Comment


                    • #20
                      Originally posted by Etherman View Post
                      Well Google makes phone software and web browsers.
                      Their web browser have no competition (I'm using Firefox again for several months, and ...just waiting to get rid of this crap), they basically are currently in former M$ IE position, they don't have any Chrome-related scheduler business here. This is already total domination and web standards dictatorship. And web browser usage isn't system exclusive - there are some background tasks, file managers, documents, media playbacks, you name it.
                      You seem to be thinking that sched_ext will give you some more performance in general, while in fact this "game tuning" simply starves non-foreground tasks.

                      As for phone software - fortunately they don't manufacture hardware...
                      Disregarding Android being Linux kernel overlayed with their custom environment controlling all the apps and hardware... and already having various game boosters.

                      Because CPU-hogging software vendor (like Valve) could be interested in 10%, or 20%, or maybe 50% performance gains, but with moderate expenses - after all, the hardware is yours and you can simply buy better, faster, more powerful one... if you can afford; otherwise it's your problem.

                      Meanwhile hyperscalers would chase even 0.1% of their electricity bills. I guess they would save zillions of $$$ by simply soft-suspending idle machines.

                      If only Google were selling hardware, they could have been tempted to make people buy newer ones every few years...


                      Of course I see a selling point here - finally Linux could deprioritize minimized windows. But appropriate Wayland protocol would take probably like 2 decades to be acknowledged...

                      Comment

                      Working...
                      X