Re: ULE vs. 4BSD in RELENG_7

Jeff,

I just ran through some of my benchmarks on a kernel build from
sources as of today, and I've noticed an improvement for the ffmpeg
workload. Here's a comparison of 4bsd, ule (BETA1) and ule (BETA3).
This is vanilla source with no patches applied:

x 4bsd
+ ule
* uleb3
+--------------------------------------------------------------------------------------------------+
|x x x x * * * * *
+ + ++ +|
| |____A_____| |_______AM_____|
|______A_______| |
+--------------------------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 5 93.28 94.05 93.74 93.746 0.29686697
+ 5 97.56 98.7 98.03 98.042 0.41924933
Difference at 95.0% confidence
4.296 +/- 0.529778
4.5826% +/- 0.565121%
(Student's t, pooled s = 0.363249)
* 5 95.58 96.69 96.37 96.278 0.42446437
Difference at 95.0% confidence
2.532 +/- 0.534176
2.70092% +/- 0.569812%
(Student's t, pooled s = 0.366265)

Unfortunately, I only have 3 data points for the buildworld and
sysbench runs. buildworld has slowed down slightly, and sysbench has
slightly improved:

buildworld:

x 4bsd
+ ule
* uleb3
+--------------------------------------------------------------------------------------------------+
|x x x xx +
+ + + +* * *|
| |____________MA_____________|
|________A________| |__MA____||
+--------------------------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 5 804.27 813.06 808.75 808.976 3.9655807
+ 5 822.11 828.74 825.13 825.248 2.5607655
Difference at 95.0% confidence
16.272 +/- 4.86816
2.01143% +/- 0.601768%
(Student's t, pooled s = 3.33791)
* 3 829.15 831.33 829.73 830.07 1.1290704
Difference at 95.0% confidence
21.094 +/- 5.90231
2.60749% +/- 0.729603%
(Student's t, pooled s = 3.30285)

sysbench (8 threads):

x 4bsd
+ ule
* uleb3
+--------------------------------------------------------------------------------------------------+
|x x xx x +
++ * + * *|
| |___A_M__|
|______AM_|____|MA______||
+--------------------------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 5 2151.75 2172.05 2166.7 2163.14 8.1262968
+ 5 2274.13 2309.66 2298.99 2296.89 13.648515
Difference at 95.0% confidence
133.75 +/- 16.3813
6.18314% +/- 0.757294%
(Student's t, pooled s = 11.2321)
* 3 2303.61 2327.29 2313.17 2314.69 11.912951
Difference at 95.0% confidence
151.55 +/- 17.0782
7.00602% +/- 0.789509%
(Student's t, pooled s = 9.5567)


Thanks,
Josh
_______________________________________________
freebsd-performance@xxxxxxxxxxx mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-performance
To unsubscribe, send any mail to "freebsd-performance-unsubscribe@xxxxxxxxxxx"

Relevant Pages

  • Re: ULE vs. 4BSD in RELENG_7
    ... ule and ule. ... This is vanilla source with no patches applied: ... Here are the ministat outputs for ffmpeg and sysbench, ...
    (freebsd-performance)
  • Re: ULE vs. 4BSD in RELENG_7
    ... ule (BETA1) and ule. ... Here are the ministat outputs for ffmpeg and sysbench, ...
    (freebsd-performance)
  • Re: ULE vs. 4BSD in RELENG_7
    ... significantly faster in ULE. ... seemed to be very bad under ULE only with 2 CPUs. ... reporting ULE had more idle time, which looks like a bug since I would expect it be basically 0% idle on such a workload. ...
    (freebsd-performance)
  • Re: ULE vs. 4BSD in RELENG_7
    ... Anyway, in summary, ULE is about 5-6 % slower than 4BSD for two ... help indicate where the deficiency is in the ULE scheduler? ... I'm not sure if Remy subscribes to this list, ... One major difference is that your workload is 100% user. ...
    (freebsd-performance)
  • Re: ULE and current.
    ... Haven't been running ULE for a while, because I was till having slight ... overall interactivity feels slightly ... hardware: dual p3-450 ... workload: general gnome desktop + kernel build ...
    (freebsd-current)