2016-01-07 06:43 AM
Hi everyone, this will likely be an easy question I just want to get confirmation before I go through the motions.
I've got 20 1.2TB SAS disks that I want to put into a new aggr. Am I better off splitting the aggregate into, for example, 2 raidgroups of 9 (2 parity, 7 data and leaving 1 spare per group) or can I just as well create the aggregate with a single raid group of 19 (2 parity, 17 data leaving a single spare)?
I know it's possible to do the latter I am just wondering if it is advisable to have an aggregate of that size with a single raid group. On the other hand, if I split it into 2 groups it isn't getting me nearly as much usable storage.
Thanks in advance.
Solved! SEE THE SOLUTION
2016-01-07 06:48 AM
What type of controller? Is this an HA config?
What is your workload going on here? do you have flashcache card
Those are all questions we need to help answer your questions.
I would tend to make a 18 disk rg with 2 spares..
2016-01-07 06:58 AM
This is a FAS3220 HA, 8.2.4, FCP, running VMFS luns for our VMWare infrastructure, no flashcache cards.
This DS2246 disk shelf was purchased recently because I've been seeing disk latency on the 2 DS4246 shelves that are loaded with nothing but 3tb SATA drives (this was configured before my time). The new shelf was populated with the 20 aforementioned SAS disks and 4 400GB SSD. My plan has tentatively been to create a new SAS aggregate and then add the 4 SSD to one of the other SATA aggregates in order to assist with the iops of that aggregate. I've run AWA against the pre-existing aggregates and it looks like the 4 SSD I have should be sufficient for one of the aggregates. The alternnative would be to put the 4 SSD as a flashpool with the new SAS aggregate but my thoughts are that those 4 SSD would be better used in a flash pool on the SATA aggregate since it is already in need of the iops and has so much storage space.
2016-01-07 07:31 AM
Ok, you are doing all the right things... Perhaps you could have run AWA or PCS prior to purchase, where a flashcache card might have helped more..
I would in no doubt create an 18 disk rg with SAS, you will have tons of IOPS for VMS.
If you are going to make a hybrid aggr, you need spares as well, you can do raid-4 with 1 spare on top
2016-01-07 07:52 AM
I would have definitely run AWA prior to purchase unfortunately when the disk shelf was purchased we were running on 8.1.4P6. I just recently completed the upgrade to 8.2.4 because at the previous version it didn't support over 600GB for flash pool which left my SSD more or less unusable for that purpose. I don't think our NetApp Sales Engineer put a whole lot of thought into what he was selling us when he sold us this new shelf, of course I really only have myself to blame for not getting more info on what we were getting versus what we already had.
Ok, here is something else that I seem to get confilcting information on every time I bring it up. NetApp documentation says that you can mix 2 types of drives in a flash pool, SSD and SAS or even SSD and SATA (Page 4 paragraph 1 of the attached PDF says exactly that). As a recent test I took some of my spare SAS and created a hybrid aggr and then added all four of my SSD in raid-6 to the aggr as the flash pool, leaving the remaining SAS drives as spares for the SAS rg as well as for the flash pool rg. Previously when I've created an aggregate/raidgroup if I don't have any spare disks available it would give me a warning prior to creation which it did not during my test. So my thoughts are that I could still add the 4 SSD to my aggregate, in raid-6 (as recommended by netapp) with one of the HDD as a spare if needed. I have been warned against creating the flash pool in raid-4 by several people since in the case of multiple drive failure or a single drive failure combined with a write error on another disk would bring down the raidgroup which would then bring down the entire aggregate that it is attached to.
What are your thoughts? I've been mulling this over for a couple of weeks now, and can't seem to land on a decision of what to do with these SSD.
2016-01-07 08:18 AM
Are you saying that you think you can use SAS spares for your flash pool aggr?
My feeling is you will drive the IOPS fine out of your SAS spindles, and if your latency numbers show that you need SSD on sata, which would make sense, then I would stack a raid 4 with a spare if you need space.. If you are worried about failure domain, then use raid-dp.. but ssd's rebuild very fast.. If you do DP, you "should" still use a spare, as a safety, measure, so you wil be down to one writeable spindle... A high cost for your deployment...
I'm not your SE, i can only speak from experience...
2016-01-07 08:30 AM
Yes that is what I was saying. Since the flash pool can according to NetApp be a combincation of SAS and SSD my thoughts are that a SAS drive should be fine as a spare within a raid-dp group.
But you make some good points about going with the raid-4 flash pool with one SSD as spare. I would just hate to have the one in a million chance where there is some combination of failures that brings down the flash pool and then as a result the entire aggregate. Sometime this year we should be upgrading to CDoT which will be nice, but that isn't going to be for a little while.
So you think doing a raid-4 flash pool (2 data, 1 parity, 1 spare) on our existing SATA aggregate is probably a safe bet? I think it should be, I just keep getting different responses from different people which has kept me second guessing myself.
Thanks for all of your input