ZFS sub-optimal performance with default setting

[Tom Evans]

On Fri, Jan 11, 2013 at 2:43 PM, Patrick Dung <patrick_dkt at yahoo.com.hk> wrote:
>
> Hi Tom,
>
> To make it simple, I have setup in this setting:
>
> Host: Intel dual core 3Ghz CPU, RHEL 6.3 x64, RAM 8GB
>
> Freebsd 9.1 -i386 VM with these setting:
> CPU: One
> Memory: 2GB
> 5GB for OS (da1)
> 5GB for ZFS (da2), no separate ZIL
>

This is ... tight! I concur with Adam, ZFS may be of little use in
this scenario. It is not so much lack of ZIL, but your pool is made up
of part of a virtualized shared resource. If you do not have a ZIL,
then synch writes are constrained by the speed of your slowest/only
disk, and I think you are at its limit.

> Installed software:
> Postgresql 9.2.2 (compile from ports) /usr/local/pgsql is a ZFS dataset
> OTRS 3.1.6 (compile from ports)
> Apache 2 install from packages
>
> zfs/postgresql/otrs/apache is in default setting, except I have turned off atime in ZFS.
>
> I have run OTRS benchmark twice, below is the result:
> Insert Time:     10000     12 s :-(     Should not take more than 5's on an average system.
> Update Time:     10000     7 s     Ok
>
> Select Time:     10000     3 s :-)     Looks fine!
> Delete Time:     10000     2 s :-)     Looks fine!
>
> Thanks,
> Patrick
>

Does performance significantly increase if you use UFS instead?

Cheers

Tom