[OmniOS-discuss] Fwd: All SSD pool advice

[Günther Alka]

SSD only pools can give you a similar sequential performance like good SAS disks but while oops of a regular disk is a few hundreds, a SSD is at a few thousands so there may be a huge improvement with iops sensitive use cases.

With ZeusRam and SAS disks, it seems a production machine.
For SSD only I would consider

- with an expander stay with SAS, without expander Sata is much cheaper
- prefer enterprise SSDs with powerless protection and a build in over provisioning
like Intel series 3500 - 3700 and newer

Desktop SSDs come without over provisioning (you may add your own) but without powerless protection (okay, your SAS disks does no have as well)

- as SSDs have that many iops, you may use Raid-Z (1-3) instead of multiple mirrors (reduce cost)

- a L2Arc is not needed but a ZeusRAM as dedicated ZIL makes a lot of sense as it is much faster than your SSD pool (and you do not need to write data twice (sync log and cached write) to your pool


Gea

> 
> 
> 
>> Am 04.04.2015 um 21:07 schrieb Chris Nagele <nagele at wildbit.com <mailto:nagele at wildbit.com>>:
>> 
>> We've been running a few 4U Supermicro servers using ZeusRAM for zil and SSDs for L2. The main disks are regular 1TB SAS. 
>> 
>> I'm considering moving to all SSD since the pricing has dropped so much. What things should I know or do when moving to all SSD pools? I'm assuming I don't need L2 and that I should keep the ZeusRAM. Should I only use certain types of SSDs? 
>> 
>> Thanks,
>> Chris
>> 
>> 
>> -- 
>> 
>> Chris Nagele
>> Co-founder, Wildbit <http://wildbit.com/>
>> Beanstalk <http://beanstalkapp.com/>, Postmark <http://postmarkapp.com/>, dploy.io <http://dploy.io/>
>> _______________________________________________
>> OmniOS-discuss mailing list
>> OmniOS-discuss at lists.omniti.com <mailto:OmniOS-discuss at lists.omniti.com>
>> http://lists.omniti.com/mailman/listinfo/omnios-discuss
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://omniosce.org/ml-archive/attachments/20150404/9cd0d802/attachment.html>