Well over 11,500 IOPS per array?
So let's see what that means:
1 disk (spindle) can do roughly 110 IOPS (9ms seek+deliver time in 1000ms per second), thus their array must have at least 105 disks. 105 * 320 GB disks is what 33600GB = 33TB... Well short of 50TB raw capacity???!!!
OTOH, 50TB of 320GB spindles is 157 spindles. Which in turn deliver together some 17,270 IOPS... Why is 11,500 IOPS being used as the benchmark?
Detecting failed disk areas is normally done in sectors. Bad sectors at that. And it is quite normal for a SMART disk to tell its handler where the bad sectors reside. (Linux' smartd does that for your home PC...) Working around bad sectors isn't that difficult. Even Windows 98 could do that... Or MS-DOS, for that matter.
What if the failed disk area is actually one of the heads or the spindle itself? That would call for replacement, for sure.
If Ataro wants their claim to fame to work, they better open up their intellectual property sharing so as for professionals to determine how (technically) sensible this marketing stuff actually is.
I'm not saying it can't be done, we're working machines (they do what we tell them) here after all. But it would take some proper convincing that it actually has been done.