2017-08-24 04:00 PM
I'm eagerly awaiting our new FAS2620. It comes with 4 SSD drives and 8x 4TB NL-SAS drives in the controller shelf, and 12x 4TB NL-SAS on an additional shelf.
I intend to run Ontap 9.2. We would like to run active/passive with all 20 NL-SAS drives assigned to one node. With root-data ADP, I was wondering how this works if 4 of the 12 internal first-shelf disks are taken up with SSD drives. Are 8 internal NL-SAS drives enough for active/passive ADP, or will the system allow me to include some of the second shelf in the ADP root-data partitions? If not, does it make sense to move the SSD to the second shelf so the first shelf can be completely populated by 4TB NL-SAS?
Solved! SEE THE SOLUTION
2017-08-24 05:54 PM
8 internal drives are enough for root-data partitioning - three root partitions will be used for each controller's root aggregate, with one spare per controller, while all the data partitions will be used for the data aggregate - and it is supported to mix the ~3.8TB data partitions with the unpartitioned 4TB NL-SAS drives in the same aggregate.
This documentation section will be of help to you and your VAR for assigning the system as active-passive
2017-08-25 11:47 AM
8 HDD's is the minimum, but you'll have a better layout with root/data partitions if you do put the SSDs in the external shelf. This will allow root aggregates to spread out to 3d + 2p + 1s partitions on each node.
2017-08-25 01:40 PM
Thanks Andris. I was wondering about that. We're coming from a VNX5200, so I'm hopeful active/passive will work out for us with the increased horepower of the fas2600 compared to the FAS2500 and our 2TB usable of flash pool. We don't push harder than about 1500 iops at any time at this point. We're using VEEAM for backup, and I intend to backup from our secondary netapp so that should distributed load.
2017-08-25 02:31 PM
The doc link Alex pointed you to will get you going for active/passive in short order.
If you're not dealing with multiple workloads and you don't have really high network I/O demands, active/passive is a reasonable choice. Even when active/active, you don't want to have either node handle much more than 50% of the total workloads, anyway (if you want acceptable performance when a node is down).