ZFS - benchmark & tuning before and after doubling RAM

Mehmet Erol Sanliturk m.e.sanliturk at gmail.com
Sat Jan 8 23:47:22 UTC 2011


On Sat, Jan 8, 2011 at 6:14 PM, Dan Langille <dan at langille.org> wrote:

> On 1/8/2011 4:33 PM, Mehmet Erol Sanliturk wrote:
>
>>
>>
>> On Sat, Jan 8, 2011 at 3:37 PM, Dan Langille <dan at langille.org
>> <mailto:dan at langille.org>> wrote:
>>
>>    I've been running a ZFS array for about 10 months on a system with
>>    4GB of RAM.  I'm about to add another 4GB of RAM.
>>
>>    I think this might be an opportune time to run some simple
>>    benchmarks and do some tuning.  Getting more out of the system is
>>    not a priority for me.  It does what I need now.  However, I do see
>>    some merit in writing something up for others to see/follow/learn.
>>
>>    The system is running FreeBSD 8.2-PRERELEASE #1: Tue Nov 30 22:07:59
>>    EST 2010 on a 64 bit box.  The ZFS array consists of 7x2TB commodity
>>    drives on two SiI3124 SATA controllers.  The OS runs off a gmirror
>>    RAID-1.
>>
>>    More details here: http://www.freebsddiary.org/zfs-benchmark.php
>>
>>    First, up, I've done a simple bonnie++ benchmark before I add more
>>    RAM.  I ran this on two different datasets; one with compression
>>    enabled, one without.
>>
>>    If anyone has suggestions for various tests, option settings, etc,
>>    I'm happy to run them and include the results.  We have lots of time
>>    to play with this.
>>
>>    --
>>    Dan Langille - http://langille.org/
>>
>>
>>
>>
>> I think , you know the following pages :
>>
>> http://hub.opensolaris.org/bin/view/Community+Group+zfs/zfstestsuite
>> http://dlc.sun.com/osol/test/downloads/current/
>> http://hub.opensolaris.org/bin/view/Community+Group+testing/testsuites
>> http://hub.opensolaris.org/bin/view/Community+Group+testing/zones
>>
>> Some of the links may disappear spontaneously because of restructuring
>> of their respective sites .
>>
>
> Looking briefly, them seen to be more aimed at regression testing than a
> benchmark.  They all seem to be the same thing (just different instances).
>
> Perhaps I am mistaken, but I will look closer.
>
>
> --
> Dan Langille - http://langille.org/
>


Please , you may assume that , you are testing whether a change in hardware
is breaking anything or not , before starting benchmarks .

After assuring that anything is not broken , your benchmarks results will be
more reliable .

The benchmarks may not test all or some required features .

Thank you very much .

Mehmet Erol Sanliturk


More information about the freebsd-stable mailing list