Re: Status of MySQL on 7 w/o patches applied

Hi,

Attilio Rao wrote:
> 2007/10/16, Peter Schuller <peter.schuller_at_infidyne.com>:
>   
>>>> Just in case you missed some debugging knobs: can you retest with
>>>> RELENG_7? (not -current).
>>>>         
>> I will, but I did disable INVARIANTS/WITNESS. I was afraid I had
>> missed something else that I am not aware of; diffing the kernel conf
>> did not indicate this was the case.
>>     
>
> Recent NetBSD benchmarks, showed that malloc debugging is a very big
> bottleneck for sybench, so you should remove it as well.
> Also, benchmarks posted where just in regard of cached datas, so
> without any need to do I/O from disks.
>
> Attilio
>
>
>   
If you run real RELENG_7 it have to be without debug in kernel, and in 
malloc,
nothing really to tune here :)

Just update your source tree to the latest RELENG_7 recompile and etc.

http://lists.freebsd.org/pipermail/freebsd-performance/2007-May/002664.html 
- here are my results with mysql with FreeBSD 6.2,
you can use it to see what I did and when I reach best performance.

 From what I remember to achieve good results you have to:
1) cp /usr/local/share/mysql/my-large.cnf /var/db/mysql/my.cnf
(or use huge if you have more one dual core CPUs)
2) sysctl kern.timecounter.hardware=TSC
3) changing default thread lib, but in RELENG_7 the fastest lib is used 
by default

compile 2 kernels with scheduler 4BSD and ULE and compare results :)

P.S. Do not forget to use the same version and compile options for mysql.

-- 

Best Wishes,
Stefan Lambrev
ICQ# 24134177