[OpenIndiana-discuss] Does the latest OI version allow ashift=12 for zpool creation?

[jason matthews]

I think I provided you the means to do that. Once upon a time there was 
a patch for zpool(1) to create ashit modifications. As far as I know, it 
never made it into an OI distribution. At some point in time  Wilson 
added this functionality to the sd driver and no one looked back.

I am interested to know what you think about the performance of your 
256b drives working in a 4k zpool. Please let me know how that works out.

I use 8k sector sizes on our SSDs (which internally do 8k writes) with 
128k record sizes for what are natively 8k writes from postgres. The 
SSDs have a mix of firmware that reports 256b and 4k sector sizes to the 
host. Nailing the sector size allowed me to skip flashing the firmware 
on hundreds of DC S3700 drives. Conventional wisdom is to align your 
recordsize with your I/O patter. However, using 8k recordsizes horribly 
fragmented the pool over time and performance plummeted.  I consider 
this formulation secret sauce for very active databases.

j.

On 10/28/15 4:40 AM, andy thomas wrote:
> What I would like to do is create a new pool with ashift=12 even 
> though the current disks have 512 byte/sectors. So when the next disk 
> fails, I can get a current model disk with 4k sectors to replace it. 
> Configuring ashift sizes at pool creation time can be done in FreeBSD 
> and (I think) in ZFSonLinux as well.
>
> Andy
>
> On Tue, 27 Oct 2015, jason matthews wrote:
>
>>
>>
>>
>> If your new drives are misrepresenting their sector size you can 
>> override the sector size, thanks to george wilson, in sd.conf. this 
>> is common for SSDs.
>>
>> http://wiki.illumos.org/display/illumos/List+of+sd-config-list+entries+for+Advanced-Format+drives 
>>
>>
>> You can gather the data for the identifying text from iostat -En or 
>> format -> select disk -> inquiry
>>
>> once the zpool created you can verify with zdb -vv <zpool> |grep ashfit
>>
>>
>> j.
>>
>> On 10/27/15 11:17 AM, andy thomas wrote:
>>> I admit I haven't been on this forum for years, such is the 
>>> reliability of my OI 148 server built in 2011. I'm using 3 x 2 TB 
>>> WD2002FAEX 512-byte sector disks for a ZFS RAIDz1 pool in this and 
>>> about a year ago, one of these disks failed - I tried fitting a more 
>>> recent WD2002 disk with 4k sectors but zpool replace complained 
>>> about the sector mismatch and in the end I had to find another 
>>> WD2002FAEX (and it was quite expensive too).
>>>
>>> This server is now free to be rebuilt so I upgraded it today to OI 
>>> 151a9 but the latest zpool doesn't seem to offer the option to set 
>>> the ashift value. I wondered if I installed the latest OI from DVD, 
>>> destroying the original pool, would zpool default to using 
>>> ashift=12? This would allow me to use both 512 and 4096 byte sectors 
>>> in vdevs.
>>>
>>> Andy
>>>
>>> _______________________________________________
>>> openindiana-discuss mailing list
>>> openindiana-discuss at openindiana.org
>>> http://openindiana.org/mailman/listinfo/openindiana-discuss
>>>
>>
>>
>> _______________________________________________
>> openindiana-discuss mailing list
>> openindiana-discuss at openindiana.org
>> http://openindiana.org/mailman/listinfo/openindiana-discuss
>>
>
>
> ----------------------------
> Andy Thomas,
> Time Domain Systems
>
> Tel: +44 (0)7866 556626
> Fax: +44 (0)20 8372 2582
> http://www.time-domain.co.uk
>
> _______________________________________________
> openindiana-discuss mailing list
> openindiana-discuss at openindiana.org
> http://openindiana.org/mailman/listinfo/openindiana-discuss
>