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 this link to view the tabled results for the IOMeter: Database pattern.
We will build diagrams for request queue depths of 1, 16 and 256.
Western Digital’s new HDDs are beyond competition at the shortest queue depth. They can only be challenged by the Hitachi 7K2000 in terms of deferred writing efficiency, yet the latter’s high response time prevents it from offering any serious competition. The new Seagate does not impress at all, competing only with the WD Caviar Green and falling behind the Barracuda LP at reading.
When the queue grows longer, WD’s pair goes ahead, boasting highly effective request reordering algorithms. The Hitachi is the only more or less worthy opponent. The Seagate Barracuda XT is mediocre and its slower cousin Barracuda LP shows no deferred writing at all.
Take note that there is a performance slump on the Marvell 9123 controller at a certain load. Hopefully, this is just a single problem.
Western Digital’s drives are still very good at the maximum queue depth but their performance is lowering as the share of writes increases. It looks like the HDDs allot all their resources to read requests, nearly ignoring write requests altogether. It is only when there are no reads in the queue left that they wake up and enable their highly effective deferred writing algorithms. Whatever the reason, they fall behind the Hitachi at high percentages of writes due to this behavior.
The Seagate XT still cannot get any of our praise. On one hand, it is more effective than the Hitachi at reading (but far from matching the performance of WD’s products). But on the other hand, its deferred writing is not very effective. All of this refers to the ICH7 controller. When connected to the Marvell 9123, the HDD is as poor as 5400rpm models.
Winding up this part of our tests, we will build diagrams showing each drive’s performance at five different queue depths.
The Hitachi has everything – deferred writing, request reordering and all – but it might be more effective at reordering requests at long queue depths.
The Seagate Barracuda LP shows very modest, nearly absent, deferred writing and its results at queue depths of 16 and 4 requests only differ in the case of pure random reading. Interestingly, this HDD behaved differently in the last test on the Promise controller. Its performance was better at writing, for example. This is a vivid example of the effect that the disk controller’s driver can make.
The Barracuda XT is influenced by the controller, too. When connected to the ICH7, this is a nice HDD with neat request reordering algorithms and modest deferred writing. Its high response time spoils its performance the most.
The Marvell controller adds a few unpleasant touches into this picture. First, it does not want to increase performance at long queue depths, which is a controller’s fault indeed. Second, we see sudden performance slumps in each graphs which must be the result of some conflicts between the controller’s and the HDD’s algorithms.
The graphs of Western Digital’s Caviar Black and RE4 resemble the Caviar Green’s but go higher and are stretched out vertically. So, we guess that Western Digital is still using its unified and highly effective firmware. The new products boast impressive NCQ algorithms which seem to conflict with deferred writing at long queue depths. For example, at a queue depth of 16 requests we can already see a performance hit at 90% writes. Interestingly, the Caviar Green holds on longer although it has a smaller cache.
The disk controller’s influence on performance can be observed again. On the Promise controller the Caviar Green had no performance hits at 80-90% writes (we use one and the same disk with the same firmware). It is because of such variations that we use the same testbed and methods to test all hard disk drives in our labs.