I’m probably displaying my ignorance here, but so be it. A standard 3u NetApp disk shelf contains 15 sleds. NetApp’s recommended size for an aggregate in normal (basic, non-fancy) circumstances is 14 disks. That means 12 data disks, 2 for parity, and one hot spare in the shelf. To me this feels both rock solid reliable and fairly inefficient for any scenarios with a large number of writes. It means every time you have a write you need to read from *12 disks twice* in order to get your parity data. Wikipedia suggests that something called WAFL deals with this problem by “coalescing writes in fewer stripes.” I don’t really know what that means and would love to be edumacated…