Shiny Space Utilization Post

A colleague: http://recoverymonkey.net/wordpress/2010/05/07/netapp-usable-space-beyond-the-fud/ has posted about space utilization; i.e. the amount of terabytes of spinny you must buy to get 'n' terabytes of LUN.

(Staggering over-simplification follows.....)
An example goes something like this:

With NetApp you get 11 1TB disks and that gives you 9TB of space for data and 540 IOPS. Only it doesn't.
With "others" you buy 18 disks and that gives you 9TB of disk space for the same IOPS (RAID10). You can also buy for RAID5 (10 disks) but you get nowhere near the same number of IOPS and therefore need to beef up the disk count by an undetermined number. Safe to say, RAID5 requires far more than the original 10 disks for 9TB visible. So unless you go RAID0 or implement RAID6 there's simply no way to get a better utilization ratio than RAID-DP - as an up-front raw number.

Now, remember that a 450GB disk isn't 450GB. At best it's 438, in Base2 it's 402GB. Ahh, but it's not. All that pesky 1000/1024 marketing masks reality. All manufacturers and storage vendors ensure that additional space is held back to account for physical blocks (all of which are 4KB chunks) that go bad. If you have one bad block the block is marked bad and a block out of the reservation is activated and the data put back there from parity (no, that's not exactly what happens but you're Windows people and the explanation is, to quote the Wicked witch from CA, perfect enough). That brings the number down some more from 402, usually by about 10%. So you're "450" disk is actually about 360. And before all you DAS jihadists go off on one, it's a disk. All the vendors do the same. Do this on DAS and the numbers are the same. And if the numbers aren't the same you're, by definition, sacrificing resiliency for capacity - and that's bad.

Go read Dimitris. Excellent and cool (if you love this kind of stuff)