FreBSD 9.1 and ZFS v28 performances

[Davide D'Amico]

Il 20.03.2013 02:37 Bob Friesenhahn ha scritto:
> On Mon, 18 Mar 2013, Steven Hartland wrote:
>> 
>> Oh and another thing if this is mysql did you set the right settings
>> for your ZFS volume e.g.
>> zfs set atime=off tank
>> zfs create tank/mysql
>> zfs set recordsize=16k tank/mysql
> 
> Very importantly, the recordsize should be set before first creating
> the database file.  The recordsize becomes a property of the file.
> Even if one sets it to 16k, the file will continue to use 128k if that
> was the setting when it was created.


To recap, a collegue of mine was able to reach the same performances of 
UFS using a lot of *magic* knobs or sysctls such as 
vfs.zfs.txg.{synctime,timeout}, vfs.zfs.write_limit_override, 
txg.write_limit_override and vfs.zfs.zil_disable=0.

So, the lesson for me is: with my hardware (R720 12core, 32GB ram), 
with my raid/disks setup, with the software I have to use 
(mysql-5.6.10-ent), with the dataset/payload I have to use I think I'll 
use UFS (I have a lot of other dbserver using UFS) or (but I will know 
this only tomorrow) I'll test CentOS on the same hardware to see how 
does it perform, so I'll post my questions and/or results on 
freebsd-performances ml :)

Thanks,
d.