We have a pair of 3040's 126.96.36.199P2 hosting roughly 5000 MSSQL (2005EE) databases. Hosts (2003EE) are fibre connected, all drives are 300GB 15k fibre drives. Netapp designed the storage with two aggregates to service SQL; logs are serviced by a 20 disk aggr (RAID DP), data/temp is serviced by a 50 disk aggr (RAID DP).
I have 5 available disks I'd like to add to the LOG aggr which is currently 20 drives, RAID DP, 8+2. My intent is to increase IOPS as much as possible. My options as I see them are as follows
1) Add the 5 drives to the aggregate. This would effectively
Add a 3rd RG
Use two of the disks will for parity
Use three disks for data drives
2) Change the RG size to 16 (or higher), and add the 5 drives to the aggr. This will effectively add 5 data drives. However I am unclear on the following:
Will the new drives be assigned round-robin across RGs, or will they all go to one RG?
Would it be better to add 4 drives in this case to maintain a balanced number of drives in each aggr?
Will changing the RG size effect current drive allocation?
It would seem option 2 would net the most IOPS, but I need to understand a little more before making a decision. Any comments greatly appreciated
I would go with second option and run volume level reallocate to distribute the existing data across all the disks evenly, however before using spare disk make sure you are adding same size disks and having at least 2 extra disks of same type left as spare in system
In my experience, a RAID size of 10 is way too conservative for RAID-DP. 16 is perfectly acceptable. If you just add the 5 drives it will add them to the last RG. You can back-fill but you'll have to do that on the command line and use the -g flag to do that. However, I would just add them to the end. With only 5 disks, it's not worth splitting them up, IMHO.