UNSOLVED
Isilon Data Protection Level
Hello,
I have a question on isilon Data Protection level. We currently have a 13 Node isilon with OneFS 7.1 and 130 TB of "physical' storage space with a Data Protection Level of +2.
With a +2 Data protection level the isilon uses just about twice the amount of space physically as we are using logically.
60 TB used logically.. 120 TB used physically.
We are thinking about changing the Protection level to +1 on the whole array. To test this we changed the data protection level on 2 shares and we gained back 50% of free space on those shares. With this test we are under the impression where we would be able to use logically 75% of the physical storage with a +1 data protection as oppose to 50% with a +2 data protection.
I was wondering if anyone has set there entire array to +1 and what are the consequences?
4119
2
Posted October 26th, 2015 17:00
+1 means protection again loss of one node OR one drive.
You will experience data loss when two drives out of your 156(?) drives fail.
Would you run a RAID 5 with 156 drives?
It seems you have mostly small files (around some 10 or 100 KB each), right?
Overhead for small files starts at +200% and drops to those 17% (depending
on the cluster size and protection setting) for larger files in the MB range.
If your small files are just "scratch" you /might/ consider protecting them at only +1,
*IF* at the same time you keep the "important" stuff on the cluster at +2 or better;
a SmartPools license would be required though.
-- Peter
4119
0
Posted October 26th, 2015 19:00
Thanks
Vince Rabuffo
NS, Systems Engineer II
Platform Engineering
NSMC Partners Healthcare Systems
Office 978-524-1140
Cell 339-221-8047
4119
1
Posted October 26th, 2015 21:00
Hi vrabuffo,
I want to echo what Peter said. What you are proposing is VERY risky.
Please involve your account SE in this discussion. Your account SE has access to the Isilon Sizing Tool and knows your workload and particular cluster setup. This would allow the SE to model some of these data protection specifics.
For your cluster, out of the 13 nodes, what node type(s) are they? How many nodepools? What kind of file size distribution? Would you be able to give us a file size histogram of the files on the cluster?
Thanks.
4119
0
Posted October 27th, 2015 05:00
Oh, right, if you have enough sub 128K files, the mirroring they invoke would certainly cause an effect like that.
Small files suck.
4119
0
Posted October 28th, 2015 09:00
per "https://support.emc.com/docu60090_OneFS_7.2.1_CLI_Administration_Guide.pdf?language=en_US&language=en_US" page 624-626:
Requested Protection settings:
Disk Space usage:
4119
0
Posted October 30th, 2015 00:00
Your physical used data may also differ from your logical used data if you have snapshots enabled - maybe there are some TB lost ?
carlilek
2 Intern
•
205 Posts
4119
1
Posted October 26th, 2015 16:00
Changing to +1 seems most hazardous to me. But I'm also confused, at n+2, you should only have 17% overhead with a homogenous 13 node cluster (see pg 241 of the web administration guide). I hate to ask, but are you sure you're not using 2x mirroring?