In the Database pattern the HDD 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% throughout the test.
The following diagrams show the dependence of speed on the ratio of reads to writes.
None of the HDDs benefits from NCQ technology. The HDDs from Western Digital and Toshiba provide the highest performance growth relative to load growth whereas the Hitachi and Fujitsu slow down when the shares of write and read requests are equal. The HDDs from Seagate and Samsung show themselves as very conservative with weak deferred writing. The Samsung is almost always below the level of 100 operations per second (except for a small range of loads at a requests queue depth of 256), showing non-aggressive look-ahead reading and a very strange behavior at 90% and 100% write requests. In the latter modes the number of operations processed by the HM160JI per second is not increasing as is the case with most HDDs (due to the deferred writing mechanisms) but diminishing. This drive seems to have non-standard firmware algorithms and you’ll see shortly if this only shows up in synthetic benchmarks such as IOMeter or in real-life tests as well.