2017-04-11 08:25 AM
I've got a FAS8020 with SAS drives and ONTAP 9.1P2. The documentation appears to imply that I can do root-data partitioning and not waste three complete drives on the root aggregate. However, much as I try, I can't seem to accomplish this.
My reference is KB 000030831:
Has anyone successfully configured root-data partitioning on a FAS system with ONTAP 9.x and SAS drives?
Solved! SEE THE SOLUTION
2017-04-11 10:36 AM - edited 2017-04-11 11:17 AM
mr kov, - you are correct, my system is not an entry-level FAS nor an AFF. (although my 8020 sure feels entry-level) However, I expect this KB to be easily updated and worded this way on purpose. Please mr. kov, put your brainpower to work on this and let's tango.
Regardless...a test is in progress, according to the cited kb. 48 drives per head are zeroing, so it'll be awhile.....
2017-04-11 10:50 PM
ADP works for small systems as there is less concurrent IO on individual drives, and for AFF systems because IO concurrency isn't a big deal in most cases. Sorry!
2017-04-12 06:43 AM
Alex - thanks for your input, and Mr. koz, thanks for your input. I wish I could declare Mr. koz's posting "the solution" again....I can't, so I gave it a "kudos".
The end result here doesn't make me happy, but I learned something else about ADP that changed my mind: If you have a disk fail that's in an ADP aggr0 /pool/partition/whatever-it-is, then you can't put the resultant spare from disk replacement back into the ADP aggr0. The ADP aggr0 takes any existing spare disk and your disk assignments for the aggr0 then become jumbled around. It seems this could be most problematic in mixed-disk systems where some are SSD and some are spinning rust.
So it looks like, as in all things, there is no free lunch.
2017-04-12 06:06 PM
It seems this could be most problematic in mixed-disk systems where some are SSD and some are spinning rust.
This goes to part of the reason why it isn't available except on all flash and low end systems, but to be clear, it's not that ADP is wild and uncontrolled in its choice of disk. With ADP, especially with ADPv2, disks aren't owned by controllers - partitions are, so yes, it does change the management paragdims a bit. With ADPv2, each device is used by both controllers in some fashion.
2017-04-12 09:14 PM
2017-04-13 12:50 PM
Following your point, mr. k.
It’s a mixed up muddled up shook up world except for....ADP....
I feel chastized and humbled. I've been around NTAP long enough, I should have known disk faults and spares would be problematic. I do expect, as we march onward arm-in-arm to Our Great Storage Future, that we will be able to move disks (be they spinning, ssd, or corn husks) into and out of service with impunity regardless of what data and metadata looks like.