[gpfsug-discuss] SSD LUN setup

[Uwe Falke]

Hi Brian, 
write endurance is one thing you need to run small IOs on on RAID5/RAID6. 

However, while SSDs are much faster than HDDs when it comes to reads, they 
are just faster when it comes to writes. 

The RMW penalty on small writes to RAID5 / RAID6 will incur a higher 
actual data write rate at your SSD devices than you see going from your OS 
/ file system to the storage. 

How much higher depends on the actual IO sizes to the RAID device related 
to your full stripe widths. Mind that the write caches on all levels will 
help here getting the the IOs larger than what the application does. 
Beyond a certain point, however, if you go to smaller and smaller IOs (in 
relation to your stripe widths) you might want to look for some other 
redundancy code than RAID5/RAID6 or related parity-using mechanisms even 
if you pay the capacity price of simple data replication (RAID1, or 3w in 
GNR). That depends of course, but is worth a consideration. 


 
Mit freundlichen Grüßen / Kind regards

 
Dr. Uwe Falke
 
IT Specialist
High Performance Computing Services / Integrated Technology Services / 
Data Center Services
-------------------------------------------------------------------------------------------------------------------------------------------
IBM Deutschland
Rathausstr. 7
09111 Chemnitz
Phone: +49 371 6978 2165
Mobile: +49 175 575 2877
E-Mail: uwefalke at de.ibm.com
-------------------------------------------------------------------------------------------------------------------------------------------
IBM Deutschland Business & Technology Services GmbH / Geschäftsführung: 
Frank Hammer, Thorsten Moehring
Sitz der Gesellschaft: Ehningen / Registergericht: Amtsgericht Stuttgart, 
HRB 17122 




From:   Brian Marshall <mimarsh2 at vt.edu>
To:     gpfsug main discussion list <gpfsug-discuss at spectrumscale.org>
Date:   07/18/2016 04:08 PM
Subject:        Re: [gpfsug-discuss] SSD LUN setup
Sent by:        gpfsug-discuss-bounces at spectrumscale.org



@Jonathan,

I completely agree on the SSD failure.  I wasn't suggesting that better 
write endurance made them impervious to failures, just that I read a few 
articles from ~3-5 years back saying that RAID5 or RAID6 would destroy 
your SSDs and have a really high probability of all SSDs failing at the 
same time as the # of writes were equal on all SSDs in the RAID group.  I 
think that's no longer the case and RAID6 on SSDs is fine.  I was looking 
for examples of what others have done:  RAID6, using GPFS data replicas, 
or some other thing I don't know about that better takes advantage of SSD 
architecture.  Background - I am a storage noob

Also is the @Jonathan proper list etiquette?

Thanks everyone to great advice I've been getting.

Thank you,
Brian

On Sun, Jul 17, 2016 at 5:49 PM, Jonathan Buzzard <jonathan at buzzard.me.uk> 
wrote:
On 17/07/16 03:56, Brian Marshall wrote:
When setting up SSDs to be used as a fast tier storage pool, are people
still doing RAID6 LUNs?  I think write endurance is good enough now that
this is no longer a big concern (maybe a small concern).  I could be 
wrong.

I have read about other products doing RAID1 with deduplication and
compression to take less than the 50% capacity hit.


There are plenty of ways in which an SSD can fail that does not involve 
problems with write endurance. The idea of using any disks in anything 
other than a test/dev GPFS file system that you simply don't care about if 
it goes belly up, that are not RAID or similarly protected is in my view 
fool hardy in the extreme.

It would be like saying that HDD's can only fail due to surface defects on 
the platers, and then getting stung when the drive motor fails or the 
drive electronics stop working or better yet the drive electrics go puff 
literately in smoke and there is scorch marks on the PCB. Or how about a 
drive firmware issue that causes them to play dead under certain work 
loads, or drive firmware issues that just cause them to die prematurely in 
large numbers.

These are all failure modes I have personally witnessed. My sample size 
for SSD's is still way to small to have seen lots of wacky failure modes, 
but I don't for one second believe that given time I won't see them.

JAB.

-- 
Jonathan A. Buzzard                 Email: jonathan (at) buzzard.me.uk
Fife, United Kingdom.
_______________________________________________
gpfsug-discuss mailing list
gpfsug-discuss at spectrumscale.org
http://gpfsug.org/mailman/listinfo/gpfsug-discuss
_______________________________________________
gpfsug-discuss mailing list
gpfsug-discuss at spectrumscale.org
http://gpfsug.org/mailman/listinfo/gpfsug-discuss