In the Database pattern the drive is processing a stream of requests to read and write 8KB random-address data blocks. The ratio of read to write requests is changing from 0% to 100% with a step of 10% throughout the test while the request queue depth varies from 1 to 256.
You can click the following link to view the tabled results in IOMeter: Database.
We will build diagrams for request queue depths of 1, 16 and 256.
The sprightly Western Digital is ahead of all under minimum load. The Samsung and Toshiba go neck and neck, the former being somewhat faster at reading and the latter, at writing. The Hitachi behaves in a curious manner: it copes well enough with lots of writes but is slow when there is the same share of reads and writes to be performed.
The Toshiba outperforms the Samsung at reading under the increased load. This seems to be due to some flaws in the Samsung’s firmware rather than to the Toshiba’s advantages. Samsung’s drive has an inexplicable performance hit at high percentages of reads. The Hitachi is again good when processing same-type requests but slows down greatly under a mixed stream of requests.
The picture changes at the highest load. This time the Hitachi is very close to the leading Western Digital at pure reading and writing but has modest performance under mixed loads again.
The following diagrams show each drive’s performance at five different queue depths.
The characteristic growth of performance in the area of low percentages of writes indicates that the Hitachi is good at request reordering. This technique shows its full effect under high loads only, though. The drive also has deferred writing, but it seems to conflict with NCQ. These features just don’t work simultaneously. The drive seems to be confused when there is about the same amount of reads and writes – and slows down under such load.
The Samsung M6S is a calm and collected device. It has very discreet deferred writing and does little of request reordering. Samsung seems to follow the cautious policy of not introducing any dramatic changes into its new products. We can see improvements over the previous models (for example, over the 250GB M5S), but server applications are clearly not for this HDD.
Here are the same diagrams for the other tested drives: