For this next set of tests, we measured access latency at various queue depths with the same fully random IOmeter 4K access pattern (67% reads, 33% writes), while each of the drives was also under a sustained sequential write workload. A sustained sequential write across the entire volume was initiated, and when performance plateaued, access latency was measured with IOMeter.
 |
|
Latency Under Load |
|
Sustained Writes w/ 100% Random 4K Access Pattern |
|
While under load, access latency increases across the board with all drives and all every queue depth. As the load on the drives increases, however, a trend emerges. The Kingston drives offered similar (though slightly better) latency at QD1, but as queue depths increased, the
Kingston drives took a larger hit and latency under load was markedly higher than the Samsung drives.
