Announcement

Collapse
No announcement yet.

BFS Scheduler v0.444 For Linux 3.12

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

  • BFS Scheduler v0.444 For Linux 3.12

    Phoronix: BFS Scheduler v0.444 For Linux 3.12

    The BFS Linux kernel scheduler by Con Kolivas is up to version 0.444 this week and this updated scheduler is available for the Linux 3.12 kernel...

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

  • #2
    Originally posted by phoronix View Post
    Phoronix: BFS Scheduler v0.444 For Linux 3.12

    The BFS Linux kernel scheduler by Con Kolivas is up to version 0.444 this week and this updated scheduler is available for the Linux 3.12 kernel...

    http://www.phoronix.com/vr.php?view=MTUzNTk
    Yes, we want to see benchmarks. Not meaningless CPU bound tasks or disk throughput benchmarks, but real world latency measurements that show the differences a gamer would experience.

    Comment


    • #3
      "Interested in seeing some fresh benchmarks of BFS?"

      Yes!

      Comment


      • #4
        As 443 was available for 3.12, too, this article is misleading. Is a bugfix really newsworthy?

        Wake me when there's BFS for 3.13

        Comment


        • #5
          I'd definitely be interested in some meaningful* benchmarks - I'm currently using BFS and it seems as if the latency is lower than a stock kernel, but knowing it's meant to be better could well be a bias.

          *As someone said above there's no point in benchmarks that test the wrong thing, and Phoronix does have a thing for meaningless ones.
          IO- or CPU-bound benchmarks would be interesting to see if there's a downside there compared to CFS, but latency is the main selling point and supposed advantage for BFS.

          Comment


          • #6
            Originally posted by FLHerne View Post
            I'd definitely be interested in some meaningful* benchmarks - I'm currently using BFS and it seems as if the latency is lower than a stock kernel, but knowing it's meant to be better could well be a bias.

            *As someone said above there's no point in benchmarks that test the wrong thing, and Phoronix does have a thing for meaningless ones.
            IO- or CPU-bound benchmarks would be interesting to see if there's a downside there compared to CFS, but latency is the main selling point and supposed advantage for BFS.
            https://rt.wiki.kernel.org/index.php/Cyclictest - go nuts ...Cyclictest is for just that (testing latency on linux).

            Here are my typical results;

            System - AMD Phenom II x4 965 (3.4ghz) / 16gig RAM + 1.5tb SATA drive / Nvidia GT240
            OS - Archlinux 64bit + linux-l-pa [ linux-rt patch + BFQ + UKSM support ] using the nvidia blob (331.20)

            results from my average Cyclictest run, under 100% load, *plus* flash vids, jackd running, etc;

            (act:/Avg:/MAX: displayed in * usec *)

            T: 0 (20548) P:95 I:1000 C: 41480 Min: 1 Act: 1 Avg: 3 Max: 29
            T: 1 (20549) P:95 I:1500 C: 27651 Min: 2 Act: 3 Avg: 2 Max: 29
            T: 2 (20550) P:95 I:2000 C: 20740 Min: 2 Act: 3 Avg: 2 Max: 19
            T: 3 (20551) P:95 I:2500 C: 16592 Min: 2 Act: 3 Avg: 2 Max: 24

            * using 0% load on (most if not all) x86_64 machines, the average MAX: will likely have higher values. (same goes if using deep-pstates, which i don't).

            Comment


            • #7
              Comprehensive benchmarks would be appreciated.

              Comment


              • #8
                Originally posted by FLHerne View Post
                I'd definitely be interested in some meaningful* benchmarks - I'm currently using BFS and it seems as if the latency is lower than a stock kernel, but knowing it's meant to be better could well be a bias.

                *As someone said above there's no point in benchmarks that test the wrong thing, and Phoronix does have a thing for meaningless ones.
                IO- or CPU-bound benchmarks would be interesting to see if there's a downside there compared to CFS, but latency is the main selling point and supposed advantage for BFS.
                By definition, just about any scheduler will lower latency over a fair scheduler, since a fair scheduler couldn't give a damn about how much an application needs to run, just that other apps haven't. For gaming, CFS is probably the worst possible scheduler you can use.

                Comment


                • #9
                  Originally posted by gamerk2 View Post
                  By definition, just about any scheduler will lower latency over a fair scheduler, since a fair scheduler couldn't give a damn about how much an application needs to run, just that other apps haven't. For gaming, CFS is probably the worst possible scheduler you can use.
                  I don't think that's entirely correct, despite the name. CFS actually handles latency better than previous schedulers such as the old O(1). It has a fair design, sure, but overall latency is relatively good. Whether it is better or worse than BFS is another matter. Certainly comparisons between older versions of CFS and BFS had BFS winning but the improvements appear to be smaller in newer releases. Good benchmarks would certainly help.

                  Comment


                  • #10
                    Yes interested

                    In particular, I'm interested how it affects latency under heavy load for both user responsiveness and GUI draws per second.

                    I actually stopped using BFS in kernel 3.12 because of the previous hibernaton/sleep problems (this is a workstation laptop I depend on).

                    Honestly, CFS seems fine, until I am trying to run Splunk (with the unix plugin) on it in the background which uses 100% of my 4 virtual cores (it would use more if it could). BFS did much better there.

                    Thanks

                    Comment

                    Working...
                    X