Benchmark Results: 4 KB and 512 KB Random Reads
The delta between these two drives is relatively small with a queue depth of one—expected given the fact that we’re hardly exploiting the parallelism of a modern high-end SSD’s architecture. At the other extreme, pushing more concurrent requests bolsters performance across the board. However, the fact that we’re using 4 KB blocks of data is what really hammers the new Vertex 2 and its 25 nm NAND.
My theoretical explanation makes more sense next to the following chart, with the same test using 512 KB reads. Suddenly, the overhead is nowhere near as severe, and we see the new Vertex 2 faring much better, surpassing the older drive, in fact.