mysql performance test results under FreeBSD-7.0-CURRENT

[Alexander Leidinger]

Ganbold <ganbold at micom.mng.net> wrote:

> Hi all,
>
> I did make some mysql performance tests under FreeBSD-7.0-CURRENT 
> with various
> scheduler and compile time options.
> It seems like mysql(BUILD_OPTIMIZED=yes, BUILD_STATIC=yes, 
> WITH_PROC_SCOPE_PTH=yes)-libpthread-tsc-sched_4bsd+preemption gives
> better performance.

You should test without WITNESS (and maybe without INVARIANTS), as it's a
safety net / debug feature for us developers.

> There are several things I didn't test and this leads to some questions:
>
> 1. I didn't make test with Poul-Henning's CPU accounting patch. 
> Somehow I can't apply it 
> (http://phk.freebsd.dk/patch/cpu_acct_2.patch) cleanly. Where can I 
> find latest patch?
> When this patch will be included in CURRENT?

It looks like a different version of this patch is committed.

> 3. I did make tests with default malloc in CURRENT. I'm confused what 
> malloc options should try (jemalloc? phkmalloc?) What is the default 
> malloc in CURRENT? How to use these different mallocs?

The default in current is jemalloc as it is supposed to give a lot of
benefits in the SMP case, and no no drawbacks in the UP case. The previous
version is called phkmalloc. Have a look at "man malloc.conf" for tuning
features. You can also disable some debug stuff by recompiling the world
with "-DNO_MALLOC_EXTRAS" added to your CFLAGS.

Bye,
Alexander.

-- 
http://www.Leidinger.net  Alexander @ Leidinger.net: PGP ID = B0063FE7
http://www.FreeBSD.org     netchild @ FreeBSD.org  : PGP ID = 72077137
The further away the disaster or accident occurs, the
greater the number of dead and injured required for it
to become a story.