The big drop in read performance in ATTO when the access pattern changes from 1*stripe size (64KB) to 2*stripe size (128KB) on both controllers is very odd.

The random I/O perf on the 9260 is very disappointing. Makes me think I should wait until Q1 next year and upgrade to the Areca 1800 series instead -- assuming they can solve the small-block latency issues, of course.