lkml.org 
[lkml]   [2008]   [Jan]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    From
    SubjectRe: [RFC][PATCH] per-task I/O throttling
    Date
    Peter Zijlstra wrote:
    >>
    >> Anyway, I'm wondering if it's possible (and how) to already do this with
    >> process containers...
    >
    > I think there is an IO controller somewhere based on CFQ.
    >
    > I don't like this patch, because it throttles requests/s, and that
    > doesn't say much. If a task would generate a very seeky load it could
    > still tie up the disk even with a relatively low setting.
    >

    Very true. A seeky intensive process wouldn't be limited at all. And I'm
    sure there're better ways/models to satisfy my needs.

    A suggestion (off-list) has been to try with ionice that seems to be the
    right solution to limit the I/O activity of single processes, but it
    doens't allow to define policies based on UIDs or GIDs.

    BTW I don't have any number to compare the effectiveness of the priority
    approach vs the throttling approach. Here is a very quick test made on
    my PC (not sure if glxgears is the right benchmark to evaluate
    the system responsiveness):

    >>>>>> starting: glxgears <<<<<<
    3564 frames in 5.0 seconds = 711.722 FPS
    3953 frames in 5.0 seconds = 790.598 FPS
    3969 frames in 5.0 seconds = 793.794 FPS
    >>>>>> starting: md5sum /usr/lib/* <<<<<<
    3769 frames in 5.0 seconds = 753.189 FPS
    2877 frames in 5.0 seconds = 572.843 FPS
    3481 frames in 5.0 seconds = 696.071 FPS
    3775 frames in 5.0 seconds = 751.404 FPS
    2781 frames in 5.0 seconds = 556.118 FPS
    3209 frames in 5.0 seconds = 641.064 FPS
    2843 frames in 5.0 seconds = 565.697 FPS
    >>>>>> starting: echo 100 > /proc/`pidof md5sum`/io_throttle <<<<<<
    3652 frames in 5.0 seconds = 730.253 FPS
    3669 frames in 5.0 seconds = 733.734 FPS
    3797 frames in 5.0 seconds = 759.234 FPS
    3883 frames in 5.0 seconds = 776.488 FPS
    3895 frames in 5.0 seconds = 778.868 FPS
    3845 frames in 5.0 seconds = 768.968 FPS
    3829 frames in 5.0 seconds = 765.793 FPS

    >>>>>> flush caches (/proc/sys/vm/drop_caches) <<<<<<

    >>>>>> starting: glxgears <<<<<<
    3763 frames in 5.0 seconds = 752.539 FPS
    3818 frames in 5.0 seconds = 763.483 FPS
    >>>>>> starting: ionice -c3 md5sum /usr/lib/* <<<<<<
    3443 frames in 5.0 seconds = 688.597 FPS
    3202 frames in 5.0 seconds = 640.390 FPS
    3807 frames in 5.0 seconds = 761.391 FPS
    3053 frames in 5.0 seconds = 610.539 FPS
    2759 frames in 5.0 seconds = 551.790 FPS
    2975 frames in 5.0 seconds = 594.873 FPS
    2993 frames in 5.0 seconds = 596.709 FPS
    3250 frames in 5.0 seconds = 649.857 FPS
    3494 frames in 5.0 seconds = 698.688 FPS

    -Andrea


    \
     
     \ /
      Last update: 2008-01-11 16:33    [W:0.025 / U:1.500 seconds]
    ©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site