During my time in storage, the debate over what an IOP really is, has always been important. Critically so to keep vendors honest.
But lately, I have bee seeing another misuse (I think) of disk IOPS.
Software transfer commands to a device being misinterpreted as a disk physical operation.
Disk affinity and NCQ can greatly exaggerate the apparent performance of a spindle if transfers (Xfers) are mistakenly or willfully used and spoken to as Disk IOPS.
If someone told you that you could expect (or from your gathered system data) many hundreds of Disk IOPS from Any spinning disk without mind warping latency, would you question that as a specious measurement?
I would enjoy hearing what others here think about this. Do you differentiate between Xfers and Disk IOPS?
This is not in any way a vendor issue or attack, just that this is the place I know where the smart kids hang out. :)
Sent from my iPad