Announcement

Collapse
No announcement yet.

File-Systems Appear To Slowdown On Linux 3.13 Kernel

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

  • File-Systems Appear To Slowdown On Linux 3.13 Kernel

    Phoronix: File-Systems Appear To Slowdown On Linux 3.13 Kernel

    Our initial file-system testing of EXT4, XFS, Btrfs, and F2FS from the Linux 3.13 kernel appear to reveal that the performance overall is slower than when using the Linux 3.12 kernel on the same software/hardware configuration.

    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

  • #2
    Originally posted by phoronix View Post
    Phoronix: File-Systems Appear To Slowdown On Linux 3.13 Kernel

    Our initial file-system testing of EXT4, XFS, Btrfs, and F2FS from the Linux 3.13 kernel appear to reveal that the performance overall is slower than when using the Linux 3.12 kernel on the same software/hardware configuration.

    http://www.phoronix.com/vr.php?view=19449
    You did not mention we are in the middle of a big change in the block layer:
    http://phoronix.com/forums/showthrea...ue-Block-Layer

    Comment


    • #3
      13.3 has been verified for Ubuntu LTS 14.04 ... so this will get some quick attention

      That being said, those are some pretty harsh regressions for any kernel status.

      Would be very interesting to see how spinning storage does.

      As the most simple inquiry ... can you see how many cores are being utilized for IO? As alluded to by Marco, there's changes afoot to spread IO across cores. Perhaps that's where the train's leaving the tracks?

      BTRFS was supposed to have a big jump in performance so what we might be seeing is a roughly equal performance hit, with BTRFS performance increases masking some of the IO performance decreases.

      Comment


      • #4
        Originally posted by dmytty View Post
        That being said, those are some pretty harsh regressions for any kernel status.

        Would be very interesting to see how spinning storage does.

        As the most simple inquiry ... can you see how many cores are being utilized for IO? As alluded to by Marco, there's changes afoot to spread IO across cores. Perhaps that's where the train's leaving the tracks?

        BTRFS was supposed to have a big jump in performance so what we might be seeing is a roughly equal performance hit, with BTRFS performance increases masking some of the IO performance decreases.
        in my pretty basic testing i didn't have anywhere near the same performance deficit on hard-disk compared to ssd. that said my hard-disk based machine is only core2duo.

        his tests show way better performance than i was getting, although it does seem my performance has gone up a bit with rc2 compared to rc1.

        Comment


        • #5
          interactivity is way better on 3.13, so i hope they won't fix this.(chrome start time from cold reduced by a forth )

          Comment


          • #6
            I think that getting better interactive responsiveness is an acceptable trade-off for doing worse in synthetic benchmarks.

            Comment


            • #7
              I tried to put 3.13 RC3 on my Mint 16 laptop and it froze before the UI came up. Had to remove it.

              Comment


              • #8
                Some quite decent numbers from F"FS there. I'm currently evaluating filesystems for a MySQL server with an Intel S3700 SSD, although I'm running 3.12.4 on that box currently. F2FS delivers superior performance on low thread counts (4-12 threads). I'm running linkbench from Facebook, with some modifications so it writes more data (working set fits in RAM anyways).

                Comment


                • #9
                  This is serious stuff. Unacceptable.

                  Comment


                  • #10
                    Originally posted by chithanh View Post
                    I think that getting better interactive responsiveness is an acceptable trade-off for doing worse in synthetic benchmarks.
                    Totally agree. I wonder what is the status now that 3.13 has been released.

                    Comment

                    Working...
                    X