I/O size

I/O size is the amount of data that each operation performs. This is dependent on the workload type, as each application interacts with the storage system differently. The I/O size impacts directly on the previously mentioned aspects of performance.

Smaller I/O results in lower throughput, but, if done fast enough, it results in higher IOPS with lower latencies. Larger I/O, on the other hand, provides a higher throughput, but generally produces lower IOPS as fewer operations are done in the same amount of time.

There is no solid recommendation regarding I/O size. In an ideal, non-realistic world, I/O is done big enough and fast enough, which results in high throughput and high IOPS. In reality, it's either one or the other. Small I/O ends up being slow regarding throughput, but it completes fast enough so that IOPS seem higher. With big I/O, the numbers are inverted, and the throughput becomes high, but since it takes longer to complete, IOPS goes down.