2.6.5 latencytest +aa5 +mm3 as/cfq scheduler

From: Marcus Hartig (m.f.h_at_web.de)
Date: 04/09/04

  • Next message: Marcelo Tosatti: "[PATCH] drivers/char/KConfig: Remove "BROKEN_ON_SMP" tag on CYCLADES entry" 
    Date:	Fri, 09 Apr 2004 23:47:17 +0200
To: linux-kernel@vger.kernel.org

    Hello,

    I've done some benchs with:
    http://www.alsa-project.org/~iwai/alsa.html#LatencyTest

    Test condition:
       CPU load = 0.8
       RTC freq = 1024
       RTC wakeup count = 2
       Deadline = 2
       X-test = yes
       Proc-test = yes
       Disk-test = yes
       Filesize = 1000000000

    With the last stable kernel release 2.6.5 compared with -aa5 and -mm3 to
    see which scheduler/kernel has a good or low latency.
    XP Barton 2800+, 512MB PC-333, Abit NF-S-V2 nForce2, SATA Maxtor 80GB,
    ASUS 5700 FX. Fedora Core 1.91 test2, xorg-x11-0.6.6-0.2004_03_30.5, GNOME
    2.6, libata, reiserfs. Same .config as possible. All fresh booted in
    init5, latencytest-0.5.2 running in each case from an xterm.

    ---------- 2.6.5-mm3 CFQ scheduler ----------
    x11.log:
    cpu_hz = 2.08855e+09
    cpu_load=0.800000 loops per run = 645209
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.06087 ms (overrun 0)
    within 1ms = 39555, factor = 99.9798%
    within 2ms = 39563, factor = 100%
    deviation = 0.021617

    proc.log:
    cpu_hz = 2.08855e+09
    cpu_load=0.800000 loops per run = 645240
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.02978 ms (overrun 0)
    within 1ms = 15406, factor = 99.9676%
    within 2ms = 15411, factor = 100%
    deviation = 0.0219986

    diskwrite.log:
    cpu_hz = 2.08855e+09
    cpu_load=0.800000 loops per run = 645217
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.03607 ms (overrun 0)
    within 1ms = 15484, factor = 99.9548%
    within 2ms = 15491, factor = 100%
    deviation = 0.0451227

    diskcopy.log:
    cpu_hz = 2.08855e+09
    cpu_load=0.800000 loops per run = 644828
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.0494 ms (overrun 0)
    within 1ms = 53058, factor = 99.9699%
    within 2ms = 53074, factor = 100%
    deviation = 0.0380116

    diskread.log:
    cpu_hz = 2.08855e+09
    cpu_load=0.800000 loops per run = 645205
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.25948 ms (overrun 0)
    within 1ms = 37351, factor = 99.9304%
    within 2ms = 37377, factor = 100%
    deviation = 0.0339693

    ---------- 2.6.5-mm3 AS scheduler ----------
    x11.log:
    cpu_hz = 2.08832e+09
    cpu_load=0.800000 loops per run = 645104
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.05296 ms (overrun 0)
    within 1ms = 38316, factor = 99.9791%
    within 2ms = 38324, factor = 100%
    deviation = 0.0216183

    proc.log:
    cpu_hz = 2.08832e+09
    cpu_load=0.800000 loops per run = 645163
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.03901 ms (overrun 0)
    within 1ms = 15405, factor = 99.9546%
    within 2ms = 15412, factor = 100%
    deviation = 0.0223874

    diskwrite.log:
    cpu_hz = 2.08832e+09
    cpu_load=0.800000 loops per run = 645114
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.06489 ms (overrun 0)
    within 1ms = 16688, factor = 99.9521%
    within 2ms = 16696, factor = 100%
    deviation = 0.0433707

    diskcopy.log:
    cpu_hz = 2.08832e+09
    cpu_load=0.800000 loops per run = 644503
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.10258 ms (overrun 0)
    within 1ms = 38427, factor = 99.9662%
    within 2ms = 38440, factor = 100%
    deviation = 0.0456989

    diskread.log:
    cpu_hz = 2.08832e+09
    cpu_load=0.800000 loops per run = 645152
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.44847 ms (overrun 0)
    within 1ms = 34704, factor = 99.928%
    within 2ms = 34729, factor = 100%
    deviation = 0.0357967

    ---------- 2.6.5-aa5 AS scheduler ----------
    x11.log:
    cpu_hz = 2.08797e+09
    cpu_load=0.800000 loops per run = 645080
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.05035 ms (overrun 0)
    within 1ms = 39333, factor = 99.9797%
    within 2ms = 39341, factor = 100%
    deviation = 0.0215519

    proc.log:
    cpu_hz = 2.08797e+09
    cpu_load=0.800000 loops per run = 644734
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.04263 ms (overrun 0)
    within 1ms = 15399, factor = 99.9675%
    within 2ms = 15404, factor = 100%
    deviation = 0.0220961

    diskwrite.log:
    cpu_hz = 2.08797e+09
    cpu_load=0.800000 loops per run = 645054
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.08033 ms (overrun 0)
    within 1ms = 14817, factor = 99.9393%
    within 2ms = 14826, factor = 100%
    deviation = 0.0457863

    diskcopy.log:
    cpu_hz = 2.08797e+09
    cpu_load=0.800000 loops per run = 645086
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.1242 ms (overrun 0)
    within 1ms = 38657, factor = 99.9586%
    within 2ms = 38673, factor = 100%
    deviation = 0.0395391

    diskread.log:
    cpu_hz = 2.08797e+09
    cpu_load=0.800000 loops per run = 645088
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.42394 ms (overrun 0)
    within 1ms = 35125, factor = 99.926%
    within 2ms = 35151, factor = 100%
    deviation = 0.0341163

    ---------- 2.6.5 AS scheduler ----------
    x11.log:
    cpu_hz = 2.08801e+09
    cpu_load=0.800000 loops per run = 645059
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.0537 ms (overrun 0)
    within 1ms = 39353, factor = 99.9771%
    within 2ms = 39362, factor = 100%
    deviation = 0.0218243

    proc.log:
    cpu_hz = 2.08801e+09
    cpu_load=0.800000 loops per run = 645044
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.03839 ms (overrun 0)
    within 1ms = 15404, factor = 99.9416%
    within 2ms = 15413, factor = 100%
    deviation = 0.0226547

    diskwrite.log:
    cpu_hz = 2.08801e+09
    cpu_load=0.800000 loops per run = 644705
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 4.51782 ms (overrun 4)
    within 1ms = 13287, factor = 99.9248%
    within 2ms = 13293, factor = 99.9699%
    deviation = 0.0516117

    diskcopy.log:
    cpu_hz = 2.08801e+09
    cpu_load=0.800000 loops per run = 645081
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 4.09889 ms (overrun 1)
    within 1ms = 37387, factor = 99.9358%
    within 2ms = 37410, factor = 99.9973%
    deviation = 0.043017

    diskread.log:
    cpu_hz = 2.08801e+09
    cpu_load=0.800000 loops per run = 645099
    total latency = 1.953125 ms
    cpu latency = 1.562500 ms
    max diff = 3.36909 ms (overrun 0)
    within 1ms = 35446, factor = 99.9295%
    within 2ms = 35471, factor = 100%
    deviation = 0.033815
    -----------------------------------------

    The vanilla 2.6.5 has overruns here, bad against the others here.

    Happy Easter all,

    Marcus
    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

  • Next message: Marcelo Tosatti: "[PATCH] drivers/char/KConfig: Remove "BROKEN_ON_SMP" tag on CYCLADES entry"

    Relevant Pages

    • Re: 2.6.5-mjb1
      ... total latency = 1.953125 ms ... cpu latency = 1.562500 ms ... deviation = 0.0196589 ...
      (Linux-Kernel)
    • Re: 2.6.5-mjb1
      ... Now, in the second try it has the loops running as the others, I ... total latency = 1.953125 ms ... cpu latency = 1.562500 ms ... deviation = 0.0216797 ...
      (Linux-Kernel)
    • Re: XP Pro file sharing (SMB) performance problem
      ... I've narrowed down the problem to amazingly bad latency in responding ... to network requests: I've got ping times from the server with mean of ... Other machines on my network (including ... have a mean delay of 3ms with less than 0.3ms deviation. ...
      (microsoft.public.windowsxp.perform_maintain)
    • Re: DirectSound Latency
      ... > input and output and 40+ ms total latency just won't cut it. ... WDM Kernel streaming is a user mode operation. ... There are other low latency API's but they may not be available will all ... You can set frame sizes as low as 5-10ms with DirectSound but total latency ...
      (microsoft.public.win32.programmer.directx.audio)
    • Re: SNTP with 1ms of precision?
      ... The latency is corrected for. ... A query-response cycle is a back-and-forth ... The assumption is that the latency for the response (going ... are half of the total latency, ...
      (comp.protocols.time.ntp)