Re: [BENCHMARK] 2.5.69-mm3 with contest

[Jens Axboe]

On Mon, May 12 2003, Nick Piggin wrote:
> 
> 
> Con Kolivas wrote:
> 
> snip
> 
> >io_load:
> >Kernel         [runs]   Time    CPU%    Loads   LCPU%   Ratio
> >2.5.68              3   492     15.9    167.1   19.7    6.23
> >2.5.68-mm1          4   128     59.4    47.6    19.4    1.62
> >2.5.68-mm2          4   131     58.8    47.0    18.9    1.64
> >2.5.68-mm3          4   271     28.4    89.2    17.9    3.39
> >2.5.69              4   343     22.7    120.5   19.8    4.29
> >2.5.69-mm3          4   319     24.5    105.3   18.1    4.04
> >
> snip
> 
> >
> >dbench_load:
> >Kernel         [runs]   Time    CPU%    Loads   LCPU%   Ratio
> >2.5.68              3   412     18.4    5.3     47.6    5.22
> >2.5.68-mm1          4   361     21.1    5.5     54.0    4.57
> >2.5.68-mm2          4   345     22.0    4.8     49.3    4.31
> >2.5.68-mm3          4   721     10.5    6.8     33.6    9.01
> >2.5.69              4   374     20.3    5.0     48.1    4.67
> >2.5.69-mm3          4   653     11.6    6.2     34.0    8.27
> >
> >Very similar to 2.5.68-mm3
> >
> Thanks again Con. These two benchmarks especially are fairly suboptimal
> compared with the 68-mm2 days... I hope it is just the larger request queue
> size in place in the rq-dyn patch in mm. If you get some time, could you
> possibly change include/linux/blkdev.h:BLKDEV_MAX_RQ from 1024 to 128 and
> bench these two loads on that setting.

Or just wait for 2.5.70, it has rq-dyn with BLKDEV_MAX_RQ of 128.

-- 
Jens Axboe