Here are the contest results (http://contest.kolivas.net) for 2.5.38-mm3:
noload:
Kernel Time CPU Ratio
2.4.19 72.21 99% 1.00
2.5.38 73.56 99% 1.02
2.5.38-mm2 73.88 99% 1.02
2.5.38-mm3 74.10 99% 1.03
process_load:
Kernel Time CPU Ratio
2.4.19 88.02 80% 1.22
2.5.38 77.44 94% 1.07
2.5.38-mm2 77.88 94% 1.08
2.5.38-mm3 77.94 94% 1.08
io_load:
Kernel Time CPU Ratio
2.4.19 170.06 45% 2.36
2.5.38 283.27 28% 3.92
2.5.38-mm2 106 75% 1.47*
2.5.38-mm3 95.9 84% 1.33*
mem_load:
Kernel Time CPU Ratio
2.4.19 100.52 78% 1.39
2.5.38 116.94 67% 1.62
2.5.38-mm2 105 74% 1.45
2.5.38-mm3 109 73% 1.50
IO load and mem_load results are averages of n=5 runs. The difference between
io_loads of mm2 and mm3 is statistically significant (p=0.022) but not mem_load.
Con.
P.S. For those who just want more info :-P
Hardware: 1133Mhz PIII, 256Mb Ram, IDE ATA5 5400rpm HD, IO load on same disk,
reiserFS
Statistics assuming normal distribution curve:
mm2:
Hi = 126. Low = 92.1
Mean = 106.
95% confidence interval for Mean: 87.52 thru 124.6
Standard Deviation = 14.9
Median = 103.
Average Absolute Deviation from Median = 11.5
mm3:
Hi = 107. Low = 82.1
Mean = 95.9
95% confidence interval for Mean: 80.51 thru 111.4
Standard Deviation = 12.4
Median = 101.
Average Absolute Deviation from Median = 9.64
On Thu, Sep 26 2002, Con Kolivas wrote:
> io_load:
> Kernel Time CPU Ratio
> 2.4.19 170.06 45% 2.36
> 2.5.38 283.27 28% 3.92
> 2.5.38-mm2 106 75% 1.47*
> 2.5.38-mm3 95.9 84% 1.33*
this is encouraging to see, 2.5.38-mm3 has the new deadline io scheduler
and it appears to be performing well here too. I'm especially looking
forward to 2.5.38 vs 2.5.39 too, as 2.5.39 will have some deadline
updates that are not in -mm3.
--
Jens Axboe