Re: 2.6.12-rc6-mm1 & 2K lun testing

[William Lee Irwin III <wli@holomorphy.com>]

On Thu, 2005-06-16 at 15:42, William Lee Irwin III wrote:
>> It's because you're sorting on the third field of readprofile(1),
>> which is pure gibberish. Undoing this mistake will immediately
>> enlighten you.

On Thu, Jun 16, 2005 at 03:25:42PM -0700, Badari Pulavarty wrote:
> Hmm.. I was under the impression that its gives useful info ..
> Here is readprofile man-page says:
>        Print the 20 most loaded procedures:
>           readprofile | sort -nr +2 | head -20

Unfortunately it's bunk. Sorting by hits gives a much better idea
of where the time is going because it corresponds to time. That's
done with readprofile | sort -nr +0 | head -20


On Thu, 2005-06-16 at 15:42, William Lee Irwin III wrote:
>> Also, turn off slab poisoning when doing performance analyses.
> Its already off. I am not trying to compare performance here.
> I was trying to analyze VM behaviour with filesystem tests.
> (with "raw" devices, machine is perfectly happy - but with
> filesystem cache it crawls).

check_poison_obj(), which appears in your profile, exists only when
CONFIG_DEBUG_SLAB is set.


-- wli