Performance in IOMark
We use our internal IOMark tool for low-level tests. Let’s begin with sequential reading.
- Data-transfer graph for Hitachi Deskstar 7K2000
- Data-transfer graph for Seagate Barracuda LP
- Data-transfer graph for Seagate Barracuda XT
- Data-transfer graph for Western Digital Caviar Green
- Data-transfer graph for Western Digital Caviar Black
- Data-transfer graph for Western Digital RE4
Let’s compare the drives according to the speed at the beginning and end of the full-capacity partitions created on them.
Everything is neat and logical here. Three out of the four newcomers have almost identical speeds, the Hitachi 7K2000 falling behind. The latter drive has a typical ratio of speed at the beginning and end (2:1), which makes us think that it uses five lower-density platters. This is sad as we’d be much more interested in platters with a reduced diameter. The lower recording density must have helped Hitachi cut the manufacturing cost (by slackening up the quality requirements) and this must be more important for the company than an advantage in speed.
And what about cache access? That’s an interesting subject as we’ve got a HDD with a new, higher-bandwidth, interface. Let’s take a close look at the results.
Hitachi Deskstar 7K2000: work with the buffer
Let’s begin with the Hitachi 7K2000. This drive’s graph is so much different from what its predecessors showed (for example, the Deskstar 7K1000) that we suspected something was wrong with our testbed. Yet the results are correct and Hitachi’s HDDs have indeed changed their behavior dramatically. The read graph is quite a shock: instead of a smooth line that quickly rises up and then flattens out, the speed and size of the requested data block are almost directly proportional. And the speed is very low. We’d see the same graph in a random read test, but we’re reading from the cache here! Obviously, the HDD does not cache read requests at all. It is only on certain data block sizes that we see its speed jump up to an acceptable level.
Writing is better, yet far from perfect, either. We just can’t explain those wild fluctuations of speed.
Seagate Barracuda LP: work with the buffer
Seagate Barracuda XT, ICH7: work with the buffer
The Seagate Barracuda XT works with the buffer differently than its predecessor. We cannot say that it works better, though. The speed of writing large data blocks is much lower than the speed of reading them, and there are some problems with writing small data blocks. The resulting graph is rugged, with slowdowns on certain data block sizes. The speed of reading large blocks from the buffer has improved, but is not consistent, either.
Seagate Barracuda XT, Marvell 9123: work with the buffer
And here is the loudly touted SATA 600. There is indeed a performance gain as the speed of reading has grown up to 300MBps. Perhaps that’s not much, but anyway. We don’t know if the speed is limited by the Marvell 9123 or the drive’s controller but we are sure that the mainboard’s controller has a very special approach to writing requests into the HDD’s cache: the speed of writing is lower than on the ICH7. Thus, the first implementation of the new SATA version is not blameless.
We carried out one more experiment. We took an ASUS P5WDG2 WS Pro mainboard and installed an add-on card with a Marvell 9123 controller into a PCIe x1 slot. An important notice: this mainboard has PCIe version 1.0 rather than 2.0 as in newer chipsets. Then we measured the speed of working with the disk buffer again.
The results are most illustrative: the PCIe bandwidth proves to be too low and the speed of reading from the buffer plummets to 200MBps. Both graphs go lower than the graphs recorded on the ICH7. It means the new high-speed SATA interface does not enjoy universal support as yet, and you should count in all the factors.
Later on, when this review was nearly complete, we got another add-on card with support for SATA 600 and USB 3.0. It differs from the above-described one in using four PCI Express lanes. We will leave USB 3.0 for an upcoming report and will focus on the Seagate Barracuda XT here.
Well, the PCI Express bus is wide enough here to allow for a speed of 350MBps when reading from the HDD’s cache. This is even better than with the mainboard-integrated controller, which must be limited by its bus, too. The speed of writing into the buffer remains low, though. This must be a peculiarity of the current version of the Marvell 9123 driver.
Western Digital Caviar Green: work with the buffer
Western Digital Caviar Black: work with the buffer
Western Digital RE4: work with the buffer
Compared with the previous drives’ results, Western Digital’s products deliver highly consistent performance. And we must note that their results are better than those of their opponents.
After the graphs, the comparative diagram does not have much to say. You can only learn that the SATA 600 interface in the Marvell 9123 implementation indeed improves the speed of reading from the buffer but worsens the speed of writing into it. Otherwise, the HDDs are all alike. The lack of details in the diagram negates the existing differences between the tested products.