Results 1 to 5 of 5

Thread: The Latest Benchmarks Of The Linux 3.13 Kernel

  1. #1
    Join Date
    Jan 2007
    Posts
    14,629

    Default The Latest Benchmarks Of The Linux 3.13 Kernel

    Phoronix: The Latest Benchmarks Of The Linux 3.13 Kernel

    While there's already been many performance benchmarks of the Linux 3.13 kernel on Phoronix through numerous articles, following this weekend's release of Linux 3.13-rc7, I ran a fresh performance comparison...

    http://www.phoronix.com/vr.php?view=MTU2MTI

  2. #2
    Join Date
    Dec 2011
    Posts
    106

    Default

    I thought the new multiqueue block layer should have "greatly increase[d] the performance of the kernel's storage subsystem with fast devices" (source). But it's RC7 and all we've got are regressions? I think I'll stay with 3.12 for a while...

  3. #3
    Join Date
    Aug 2012
    Location
    Pennsylvania, United States
    Posts
    1,877

    Default

    Quote Originally Posted by halo9en View Post
    I thought the new multiqueue block layer should have "greatly increase[d] the performance of the kernel's storage subsystem with fast devices" (source). But it's RC7 and all we've got are regressions? I think I'll stay with 3.12 for a while...
    2 points:

    1) Regressions are bugs, they will be fixed in due time.
    2) A few users here on the forums have stated that while they haven't seen actual speed-ups in transfer speeds or the likes, they have seen a notable improvement in responsiveness under I/O load.*

    *In case you didn't know, or if anyone reading this doesn't know, one of Linux' most long standing, annoying, and unsolved problems has always been that under heavy I/O load the entire system seems to deadlock, meanwhile Windows and Mac (while they do slow down) remain fairly usable and don't appear to be 'frozen.'

    So even if this doesn't actually give us speed-ups, if this helps the responsiveness issue by taking the load off of the first core exclusively then the new multiqueue block layer is already a massive improvement

  4. #4
    Join Date
    May 2007
    Posts
    69

    Default

    Quote Originally Posted by Ericg View Post
    2 points:

    1) Regressions are bugs, they will be fixed in due time.
    2) A few users here on the forums have stated that while they haven't seen actual speed-ups in transfer speeds or the likes, they have seen a notable improvement in responsiveness under I/O load.*

    *In case you didn't know, or if anyone reading this doesn't know, one of Linux' most long standing, annoying, and unsolved problems has always been that under heavy I/O load the entire system seems to deadlock, meanwhile Windows and Mac (while they do slow down) remain fairly usable and don't appear to be 'frozen.'

    So even if this doesn't actually give us speed-ups, if this helps the responsiveness issue by taking the load off of the first core exclusively then the new multiqueue block layer is already a massive improvement
    Regressions have not always been solved during the kernel cycle, but sometimes it took 2-3 kernel later to get those regressions fixed.
    The performance penalty here, looking at those benchmark seems quite huge.
    It should be a top priority to solve a bug of this size.

    Particularly considering that kernel 3.13 will be the kernel next Ubuntu LTS (and all the derivatives) will be shipping with.

  5. #5
    Join Date
    Dec 2011
    Posts
    106

    Default

    While I'm glad about the reported "notable improvement in responsiveness under I/O load", I hope it doesn't take too long to fix the regressions. These in particular are not new to rc7 and have been around for a while.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •