Winding up this part of our tests we will build diagrams showing the performance of each HDD at five different request queue depths.
Hitachi’s new firmware is a mess. A good and steady HDD should not show such an unpredictable behavior that betrays low efficiency of request reordering and problems with mixed loads at the shortest queue depth.
Samsung HDDs have been improving: the firmware of the third generation shows some deferred writing and NCQ which we did not see in the older generations. But it is sad that those algorithms are far from aggressive and are also accompanied with obvious errors, indicated by the performance hits under certain loads.
Seagate’s products show some progress as well, if we compare the 7200.11 and 7200.12 series. If the 7200.12 had faster heads, it wouldn’t even be inferior to its predecessor at short queue depths. And it is obviously better at long queue depths.
The Seagate LP shows completely different algorithms. We don’t know the reasons, but it can reorder requests very well at pure random reading but loses this ability as soon as there are write requests in the queue. We would prefer the firmware of the 7200RPM models as more stable.
The WD 7200RPM drives split up into pairs, one of which includes two models from different series: the Caviar Black and the Caviar Blue A7. And these two deliver so similar results that we might suspect them to be one and the same HDD marked differently. The other Blue drives, M9 and V1, behave in their own way which does not really match the single difference of this pair from the first one: one rather than two platters. This Blue pair have somewhat better deferred writing algorithms but are far less effective at request reordering. This can hardly be explained by their worse response time, but perhaps they are equipped with a weaker processor and thus have different algorithms.
Things are even more complicated with the Caviar Green series: the model with 32MB cache, which is supposed to be theoretically faster, is inferior to its cousin with 16MB cache and two rather than one platter. Has the transition to higher-density platters been so difficult even at the reduced rotation speed?