2011-09-26 04:01 AM
I'm adding to an existing installation and I'd appreciate a sanity check of what I'm planning to do...
The system is a HA pair of 3140's running 8.0.2p1 with SATA DS4243 shelves. I'm adding 3x DS2246s (on new loops) which are filled with 450GB SAS drives, i.e. 72 drives total. The space is to be used for VMware virtual machines that are stressing the existing SATA disks.
I'm thinking that the drives in each new shelf should be allocated as: 1 spare for filer A, 1 spare for filer B, 11 disks to create an aggregate for filer A, 11 disks to create an aggregate for filer B.
i.e. 6 spares total (reasoning: >2 per filer for HA; layout means shelf failure still leaves 2 per filer)
6 new aggregates (3 per filer) - each entirely contained within a shelf (reasoning: small number of shelves means protecting against shelf failure is impossible with reasonably-sized aggregates, so go for "damage limitation")
Splitting each shelf between the two filers allows for sharing the storage/load equally between the two filers despite the odd number of shelves.
Is this sane? Is there a better way? (e.g. create a single 11+11+11 aggregate on each filer) Do I need to worry about raid group sizes?
Thanks for any advice or suggestions!
2011-09-26 04:29 AM
So my opinion only, but nothing you are doing is wrong…however the only thing I’d consider is that bear in mind performance is based on the amount of spindles in an aggregate, so creating lots of “smaller” ones potentially could effect your disk performance, so just be aware in terms of IOPS sizing.
You may be better assigning at least one full 2246 to each controller and either splitting the third on…and either creating a big aggregate containing 33 disks per controller, or separate aggregates.
Of course it’s a balance between performance, capacity and resilience, but if you go for one shelf to each controller and then decide on the best approach for the third you could give yourself a couple of 22 data disk aggregates (maybe take the spares from the third shelf)…
So no real right and wrongs based on what you’ve said…think about performance, capacity V resilience.
As for RAID group sizes, NetApp suggests unless there is a real requirement, go with default sizes, probably wise advice!
Assuming you’re not doing any local syncmirror/metrocluster type stuff here then? To protect you from shelf failures etc…?
Hope that helps…