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 these links to view the tabled results:
We will build diagrams for request queue depths of 1, 16 and 256.
We’ve got rather odd standings at the shortest queue depth. The Seagate Constellation ES and WD Caviar Black E8 are the leaders in terms of deferred writing. The E8 may be better than the J7 due to some internal improvements (although both use the same version of firmware), but the results of the two dual-platter models are lower than expected.
When the queue depth grows longer, the gaps between WD’s drives with different platters become larger. The newer dual-platter models are also less effective at request reordering. It is not easy to find the leader among their opponents: Seagate’s Constellation ES is only ahead at pure reading or writing. At mixed loads, the Hitachi with new firmware delivers better performance. By the way, the improvement in the latter’s firmware is obvious here.
The Samsung F3 looks best among the power-efficient products. Of course, the Seagate Barracuda LP is very effective at reordering read requests but slows down too much when there are write requests in the queue. The WD Caviar Green behaves like its senior cousins but is hamstringed by its high response time.
The results are still very interesting at the longest queue depth. The old Caviar Black drives from WD are still ahead of the dual-platter models, enjoying a large lead. The new models have smoother graphs here, though. They do not slow down that much at 90% writes. The Caviar Black E8 is competing with the Seagate Constellation ES for top place, being overall better due to higher performance under mixed loads.
As for the power-efficient products, the Samsung EcoGreen F3 loses its ground. Its performance does not scale up as much as that of its opponents at such a long request queue. The Seagate Barracuda LP, on the contrary, speeds up rapidly and takes a leading position despite its very modest deferred writing.
Winding up this part of our tests, we will show you diagrams with each drive’s performance at five different request queue depths.
Hitachi has made some obvious progress in writing firmware. The new version produces smoother graphs and is free from the problem when the HDD slows down at a longer request queue. NCQ is obvious. Deferred writing, even though not very effective, can also be seen. If these algorithms were more efficient, the HDD’s performance would scale up better at longer queue depths.
The Samsung EcoGreen F3 does not show us anything new. Its behavior is similar to the company’s earlier products and its performance is low under most loads. It is only at long queue depths that this HDD is good at reading. But it is slow at mixed loads then – slower even than at the smaller queue depths. It looks like the HDD allots all its resources to reading and there are no cache lines left for deferred writing.
Oddly enough, the power-efficient Seagate Barracuda LP and the server Constellation ES behave in a similar manner whereas the Barracuda XT is completely different. The first two models boast a highly impressive algorithm of read request reordering but slow down as soon as there are any writes to be processed. Both models also speed up at small queue depths when there are more writes than reads in the queue. It looks like the HDD is monitoring the load and decides when it is the right moment to allot its resources for writing. And there are not so many resources, by the way: they are used up even at the smallest queue depth. As a result, there is no effect from the increased queue depth at high percentages of writes.
The Barracuda XT behaves in a demure, classic way. It has all the necessary optimization algorithms but they all work with rather average efficiency.
These diagrams illustrate the changes in firmware of Western Digital’s Caviar Black series introduced together with the 500GB platters. The algorithms have changed noticeably, and not for the better as concerns server loads. The request reordering is now less effective, and the deferred writing has become less aggressive. Interestingly, the FAEX with 64MB buffer is not far better than its E3 cousin: the difference between the 3-platter E3 and E8 is larger.
We don’t quite understand the purpose of these changes. We suspect that there are fewer but larger cache lines in the newer HDDs. This should have a positive effect on performance under loads typical of desktop computers rather than servers.
When unaligned, HDDs with Advanced Format technology suffer a terrible performance hit in this test due to the lack of deferred writing. But when aligned, the WD Caviar Green with 64MB cache behaves just like other HDDs from WD. Although it does not set any records, we cannot find any real fault with it, either.