[r4] a few benchmarks

[r4] a few benchmarks

[Cyrille Chepelov]

Hi all,

I performed a couple of benchmarks on a spare machine, and put them online.

I couldn't perform all the tests I wanted (see my previous "oops" message),
the hardware is not exactly representative of today's machines, some details
would use some tuning for more representative results, and I had
ideas for extra things I'd like to measure, but here are today's results anyway
for what it's worth: http://www.chepelov.org/cyrille/fsbench-results

	-- Cyrille

-- 

Re: [r4] a few benchmarks

[Hans Reiser]

Cyrille Chepelov wrote:

>Hi all,
>
>I performed a couple of benchmarks on a spare machine, and put them online.
>
>I couldn't perform all the tests I wanted (see my previous "oops" message),
>the hardware is not exactly representative of today's machines, some details
>would use some tuning for more representative results, and I had
>ideas for extra things I'd like to measure, but here are today's results anyway
>for what it's worth: http://www.chepelov.org/cyrille/fsbench-results
>
>	-- Cyrille
>
>  
>
please look into how he measures fragmentation, reproduce his result, 
and figure out why we have any at all.  Cyrille is right that there 
should be none at all.

-- 
Hans

Re: [r4] a few benchmarks

[Hans Reiser]

Cyrille Chepelov wrote:

>Hi all,
>
>I performed a couple of benchmarks on a spare machine, and put them online.
>
>I couldn't perform all the tests I wanted (see my previous "oops" message),
>the hardware is not exactly representative of today's machines, some details
>would use some tuning for more representative results, and I had
>ideas for extra things I'd like to measure, but here are today's results anyway
>for what it's worth: http://www.chepelov.org/cyrille/fsbench-results
>
>	-- Cyrille
>
>  
>
One of the key things in reproducing these results may be the 32 meg of ram.

-- 
Hans

Re: [r4] a few benchmarks

[Mike Fedyk]

On Sat, Aug 16, 2003 at 10:30:28PM +0200, Cyrille Chepelov wrote:
> Hi all,
> 
> I performed a couple of benchmarks on a spare machine, and put them online.
> 
> I couldn't perform all the tests I wanted (see my previous "oops" message),
> the hardware is not exactly representative of today's machines, some details
> would use some tuning for more representative results, and I had
> ideas for extra things I'd like to measure, but here are today's results anyway
> for what it's worth: http://www.chepelov.org/cyrille/fsbench-results

Nice.

You should umount the partition if you want to clear the cache, and you are
doing this between filesystems so that might affect your results also
somewhat (you umounted before the mkfs for the next filesystem tested, right?)