To specifically measure latency, we use a series of 512b, 4K, and 8K measurements. At each block size, latency is measured for 100% read, 65% read/35% write, and 100% write/0% read mixes.
Normally, with single SSDs, we limit queue depth to 32, but on RAID cards with multiple drives, or other devices that can handle it, we push the queue depths much higher. In this case, we found that 384 was a sweet spot for performance.
Considering the high queue depths, the response times were great. Both 512 byte and 4K were nearly identical across the board.
The maximum latency numbers were also impressive, especially for 100% read operations. This wasn’t much of a surprise, since RAID adapters normally give good latency results. Since we had all caching disabled and the access pattern was across the entire span, you can expect even better results when all optimizations are utilized.
Now lets take a look at how consistent those results are across many hours of testing.