Well Sunnyvale obviously.
For those of us living in the reality based community and who are not charged with attempting to retcon things to suit whatever stunt marketing narrative is the current NetApp corporate necessity in the face of ever slowing revenue growth, I have to say..
Your RAID 10 baseline was stupid last week. It's still stupid this week and no matter how many days weeks or months pass between attempted justifications it will still be stupid.
Now NetApp have their own well established capacity utilization woes but recently they've taken to speaking about what they call "maximum theoretical utilization" (Which the first time I heard it sounded like the steaming pile that it turned out to be), and running comparisons against Mirrored configs.
When they say dual parity don't mention RAID 6 because they'll wave off a 14+2 (Oh look, the same number required to qualify for their Guarantee!) RAID 6 Group as something which will "significantly" reduce capacity.
In comparison to what? Mirroring? The capacity difference between a 7+1 RAID 5 group and a 14+2 RAID 6 group when you count the same number of spindles is...zip. Nada. Nothing! And both deliver more useable capacity than mirroring.
The performance impact for RAID 6 comes about in intensive WRITE situations where the I/O is random and exceeds the available back end IOPS and bandwidth. READ performance of course receives the inherent benefits of striping across multiple spindles.
Keep those in mind you'll see them again.
Lets look at their Shell Game Guarantee, the one which specifies in incredibly narrow terms that the data sets supported will not be write intensive.
Performance configs are immediately booted out.
The profile we're looking at is explicitly not WRITE intensive. No databases. No floods of random writes. We're dealing with VMs where they specify high levels of commonality to be de-duplicated it will be a READ intensive workload.
Which makes it fine and dandy for RAID 6.
When it comes to RAID types and VMware let me be clear about this by paraphrasing a response I wrote to a comment I received from a NetApp employee nearly a week ago.
On EMC gear for VMware, RAID 6/5 is absolutely fine and performs great.
EMC offers RAID 10 across all platforms and with all protocols supported (No need of a kludge like SyncMirror), to any customers who desire it or for any specific corner cases where it's a great fit.
There are a few corner cases where RAID 10 is a great fit, logs which need to be on RDMs and Enterprise Flash Drives in high end use cases spring to mind, but again those are specific corner cases.
RAID 10, 5 and 6 are in production environments today.
This Guarantee as presented, with it's drought of random writes and READ intensive workload, using a RAID 10 baseline instead of RAID 6 or RAID 5 is either an exercise in stupidity or a case of intellectual dishonesty.
I'll let the reader make up their own mind on that one, factoring in the unfortunate habit NetApp have of disabling the write cache on LUNs anytime they get in front of a Clariion.
