We have a tough decision on our hands, and I'm looking for some serious real-world experience to (hopefully) help us make a purchasing decision.
Here's a very brief overview of our environment for perspective:
4x 32-core AMD Opteron hypervisors (256GB RAM per hypervisor)
1x Dell MD3200 with 72 drives hanging off with a mix to 10K and 15K SAS6 spindles (146GB 15K for random R/W traffic, and 600GB for seq traffic)
VMWare 5.1 Standard
24x Win2k8 R2 / SQL Server 2008 Virtual Servers (each with 8 vCPUs, 32GB RAM)
So this is what I would consider a pretty typical N-tier virtualized SQL environment. We size our virtual servers basically based on portability, and the number of clients that "max out" the 8 vCPU limit on 5.1 Standard.
So far we're pretty happy with the MD3200, but there are the obvious limitations like only having 4x connected hosts in an HA config. We reached out to our IT vendor/partner regarding upgrading to something a little more "enterprise" than the MD3200, and they came back with two options:
- Dell MD3620i - Basically Dell's 10GBe iSCSI solution. Not sure if the controllers are any faster than the MD3200 (might be exactly the same).
- NetApp FAS2240 - NetApp's entry-level "FAS" array that supports all their de-dup, compression, flash pools, snapshotting, etc.
If we were running in a bare-metal SQL environment, the decision would be a no-brainer... The way NetApp "virtualizes" you storage layer makes administration and scaling of a bare-metal server environment drastically easier. Things like resizing volumes, snap-mirroring, snap-clones, etc.
However in a fully virtualized VMWare environment, there seems to be a ton of "overlap" in the features. Meaning, I can already performance snapshots which enable features like VDP, SRM, and pretty simple load-balancing through vMotion. If a particular VMFS volume is getting too busy, I can VERY easily storage vMotion that client's virtual server onto a newly provisioned VMFS or one that is less busy.
I understand that NetApp is very proud of their De-Dup engine that justify the cost increase by the cost saving in deduplication, but aren't you still cutting your raw IOPs in half? Meaning, if you can jam 50% more data into the same disks using de-dup onto 12x spindles, aren't you basically getting 50% LESS raw IOPs when compared to a solution without de-dup that uses 24x spindles?
So to those of you who have used NetApp gear under a VMWare environment, do you think it's worth the price premium? For example, NetApp works out to be almost exactly 2x the price of the Dell hardware, so if factors out to the same price. So if we're not using the premium features, it seems like a wash.
Thanks,