Winding up this part of our tests, we will build diagrams showing the performance of each HDD at five different request queue depths.
It looks like we can’t expect any serious performance benefits from Hitachi’s new firmware. These graphs are usually highly sensitive to any changes in firmware algorithms but the two sets of lines are almost identical here. The good news is that there are no such problems as we saw with Hitachi’s earlier firmware. Everything is more or less normal, although requests reordering might be more effective.
This test is illustrative of Samsung’s change of platform. It is impossible not to tell the graphs from each other. The EcoGreen F2 is dramatically different. However, we can’t say that the newer products behave in the way we like. They are not confident at high loads, suffering a performance hit at very long request queue depths. Alas, Samsung has not been able to develop server-optimized firmware. It is also sad that the HDDs do some request reordering only at very long queue depths.
Interestingly, the SpinPoint F4 is somewhat different from its Green series mates. That’s no advantage because it is just worse under mixed loads.
We don’t see anything new in the Western Digital camp. This maker’s HDDs all behave in the same way. Their graphs scale up according to how fast the read/write heads are in the specific model and how much cache memory they have. The first thing affects the height of the graphs and the second thing, how curvy they are, i.e. how effective the HDD’s deferred writing and, to a lesser degree, request reordering algorithms are. Overall, the Western Digital HDDs all perform in the same way and show the characteristic small performance hit at 90% writes and long request queue depths. People at Western Digital must be thinking that they have already reached the perfection and don’t want to change anything.
Our words do not refer to the HDD with missing deferred writing, of course. It is very poor in comparison with the others.