by Sergey Romanov
05/23/2005 | 10:11 AM
“Eureka!” he screamed out, his teeth shining through his beard.
“Gentlemen, you may congratulate me and we may congratulate each other. The problem is solved.”
”You have found a way up?”
”I venture to think so… We can at least all reach the summit,” said he.
“When we are up I may be able to show you that the resources
of an inventive mind are not yet exhausted.”
The Lost Word by Arthur Conan Doyle
They were supposed to be dead. They were a legend, an object of devotion among the initiated and a fright for the youth. The species of ten-headed colossi seemed to be doomed by Evolution itself, but they are back now!
I think this thing could only have come from the Japanese with their Godzilla and gigantic war robots. We all knew the Hitachi engineers had got hold of all the intellectual property of IBM’s disk department but none of us could ever imagine that they were going to wipe dust off the last-century drafts. The top model of the 75GXP series was the last member of the glorious breed because the production of five-platter giants was found unprofitable back in 2000. And now, four years since, we see them under the Hitachi banner. Let’s have a closer look?
I think the reasons for what has happened are rooted in the ambitions of Hitachi Global Storage that wants to be the leader on the hard disk drive market, in the slowdown of the data density growth during the last two years, and in the availability of IBM’s projects. These three components like charcoal, sulfur and saltpeter combined to make the explosion-ready Deskstar 7K400. Is it as deadly as gunpowder? Yes it is, for the competitors! Even transitioning to platters of a much higher capacity, they won’t have a drive with a larger storage space than Hitachi has now. It means Hitachi has really become a long-time leader in this parameter.
Using the same platters as in the Deskstar 7K250, Hitachi achieved the record-breaking capacity of 400 gigabytes. By the way, the problems with the data density growth have impelled other manufacturers to increase the number of platters per drive, too. For example, Seagate has mastered three-platter configurations which had been missing in the company’s product range since ancient times. But does the 7K400 differ from its predecessor in the number of platters alone? No. They also equipped the drive with new electronics that supports the UltraDMA/133 protocol. Thus, Hitachi became the third company – after Maxtor and Samsung – to embrace the innovation of the late Quantum. The redesigned electronics will also probably show some changes in the firmware, I guess. Besides that, realizing quite clearly the market prospects of the product, Hitachi equipped it with a unique – at least at the moment – feature that I will discuss at length in the next section. Right now let’s learn the characteristics of the 7K400 better.
By the way, talking about the market prospects, Hitachi intends the new drive for use in digital video-recorders and digital video processing devices in the first place. Next go various data storage and backup devices, high-performance workstations, gaming computers (a devout gamer will surely find how to use up all these 400 gigabytes), and home multimedia centers.
The low-level characteristics coincide with those of the Deskstar 7K250 (you can refer to our article called Hitachi Deskstar 7K250: Vancouver 3 HDD Review for details), with two minor exceptions: 271KB of buffer memory is now allotted for the firmware (against 254KB in the previous model) and the linear read/write speed is now higher by 100KB/s. Again, these are just minor changes. The declared average seek time hasn’t got worse despite the heavier actuator, but I am going to check this in practice soon. The power consumption has quite expectedly got higher.
But not catastrophically so: 13 watts against the 7K250’s 10 watts. Like with the previous model, the Serial ATA version has a bigger appetite on the +5V line, but the overall consumption is moderate: less than that of the voracious Barracudas on the +12V and a little higher than that of the competitors on the +5V.
The new model hasn’t got much louder. Here are the comparative acoustic characteristics of the 7K250 and 7K400:
As you can see, there’s only 1dB of difference between the three- and five-platter models. This was made possible by the improved design of the drive.
Top panels of the 7K400 (left) and 7K250 (right)
Take note of the screw that had previously disappeared on the transition to motors on liquid bearings. This screw is here again because of the new motor whose axis is fixed from two sides. Another innovation is a special latch that locks the actuator when the heads are parked on the ramp, thus preventing the heads from touching the surface of the disc at an accidental shock. This latch has been ported from the SCSI drives of the UltraStar series and will be useful for the hot-swapping feature which is declared for the Serial ATA version of the 7K400.
Another server-friendly technology is the special Rotational Vibration Safeguard (RVS) system that helps the drive to sustain vibration without losing in speed. Several hard drives in one system case always create some vibration which can knock the heads off the necessary track and bring about a considerable time loss due to the repositioning of the heads back. The electronics of the 7K400 have two acceleration sensors that keep track of vertical and horizontal vibration. The data from the sensors are used for dynamic correction of the position of the heads. This is how Hitachi estimates the efficiency of this system (the 100% point is the performance of the drive in WinBench without any vibration):
It should be noted that performance loss due to vibration may occur with a single drive, too, if it is not firmly fastened in the system case. So, this technology is undoubtedly a welcome feature. Now let’s take a look at the bottom of the drive, at its electronics.
The electronics of the 7K400 (left) and 7K250 (right)
You can see one new chip here – a TSOP-packaged chip of SDRAM memory. The memory had previously been on the reverse side of the PCB, but putting five platters into a standard-size case limited more the thickness of the electronics. The fetch time is 7.5 nanoseconds which corresponds to 133MHz frequency.
We can also see that the Serial ATA interface is still implemented by means of a translator chip, so we can forget about Serial ATA Native Command Queuing here. Instead, they have grafted a new command set for processing streaming data into the electronics.
Processing digital video in real time is different from working in an ordinary computer so far as the data storage device is concerned. Particularly, maintaining a constant data-transfer rate becomes most important. A lower speed is not permitted, but a higher speed is simply unnecessary; this mode is called isochronous. Also, more than one data stream may be required. Ordinary hard drives are not suited to function like that: the read/write speed depends on the address of the initial sector, while the data seek time as well as the time of the overall reaction of the device are not strictly regulated. That’s why the X3T13 committee that develops ATA standards came up with the Streaming Feature Command Set. The Hitachi Deskstar 7K400 is the first drive to support it.
Using the new commands, the application can get information about the speed of the drive and define the required parameters of the streams. Besides that, it is possible to prevent speed loss due to read (and write) errors since digital video playback doesn’t need 100% integrity of the data: a few incorrect bytes will only bring about barely visible distortions in a single frame, while the attempts to re-read the bad spot of the disk may lead to entire frames being lost altogether.
The first command of the new set is called Read Streaming Performance Log. It returns a series of tables with data about the number of density zones and the time it takes to read a sector in each of those zones as well as to move from one position on the disk to another one (it’s kind of the seek time parameter). This information helps the application to calculate the stream-processing strategy correctly.
The second command defines the parameters of one of eight streams, setting it for reading or writing, choosing the number of sectors per transfer and the time limit for each transfer.
The read stream command besides such traditional parameters as address and the number of sectors has the following parameters: stream ID, continuous bit (the device doesn’t process an error longer than the defined limit), not sequential bit (do not perform unnecessary look-ahead reading), urgent bit (the command should be completed in the minimum possible time regardless of errors), handle streaming error bit (process the other streams and then “reanimate” the current one).
The write stream command also has some unusual parameters: stream ID, urgent bit (perform the command in the minimum time possible), continuous bit (don’t stop on write errors), not sequential bit, and the flag of returning to the previous error in the given stream.
The last of the new commands reads the list of errors that have occurred during the processing of streams with the continuous flag set. In this mode the drive doesn’t complete the command with an error message but just signals about it with a special flag, storing the error status in a special table that can be read by this command.
Thus, the new commands offer more control over the read/write process and help to adjust flexibly for the capabilities of the given drive.
Hard disk drives with the ATA interface were benchmarked on the Promise Ultra133 TX2 controller (BIOS 126.96.36.199 and driver 188.8.131.52).
Drives with the Serial ATA interface were tested on the Promise SATA150 TX2 controller (BIOS 1.00.033, driver 184.108.40.206).
We also publish the results of a single drive attached to the Serial ATA controller integrated into the i865PE+ICH5R chipset.
For WinBench tests we created a single partition and formatted it in FAT32 or NTFS with the default size of the cluster (we use Paragon Partition Manager for FAT32 partitioning). We run each test seven times and publish the best number. The drives do not cool down between the tests. For FC-Test we created two 32GB partitions on the disk.
We use Sequential Read, Sequential Write, Database, Workstation, File Server and Web Server patterns for IOMeter tests. The File and Web Server patterns are the same as they use at StorageReview, while the Workstation pattern is of our own making, based on the statistics of access to an NTFS5 partition taken from the SR methodology. This pattern differs from the server ones in having a smaller load range and a bigger percentage of write operations (a serious workstation is supposed to have a large memory that helps to cache read operations effectively).
You can check our article called Sil 3114 SATALink 4-Port PCI Host SATA RAID Controller from Silicon Image for a thorough description of the patterns.
First, I’m going to check the Deskstar 7K400 with our new tools from the IOMark suite.
For me, it was a kind of revelation that the Serial ATA and Parallel ATA versions of the Deskstar 7K400 use firmware with different look-ahead reading! That is, they didn’t equip the HDS724040KLAT80 with the adaptive algorithms we could see in all Deskstar 7K250 models with an 8MB buffer (see our article called Hitachi Deskstar 7K250: Vancouver 3 HDD Review). This drive just got classic look-ahead reading that traces its origin back to the Vancouver2, but somewhat less aggressive and segmented. We’ll see in the following tests how bad this is. Right now let’s analyze the rest of the data.
The segmentation of the Deskstar 7K400 is somewhat smaller than that of the previous models. As you know from our Vancouver3 review, Hitachi drew a separating line between senior and junior models by means of segmentation, but this separation was rather vague in some cases. Thus, there can be situations with the 7K400 losing due to the lesser flexibility of the look-ahead algorithms, but the probability of this is very small. By the way, Hitachi still declares up to 128 buffer segments at reading, but we still don’t see that in reality.
Despite the smaller total volume of the buffer (more memory is allotted for the firmware), the read buffer size of the 7K400 has become more than 100 kilobytes bigger. But it seems like only the Serial ATA models, with adaptive look-ahead reading, are going to profit by that. Another consequence of the growth of the read buffer may be a reduction of the deferred write buffer. I can’t yet check this out, but I will keep this possibility in mind when analyzing the results of the tests that depend on deferred writing.
The average access time (the AAT column of the table) has worsened by a negligible fraction of millisecond against the previous models. Moreover, the average seek time (the AST column) fully conforms to what the manufacturer promised.
Against our tradition, I am going to begin this review with WinBench rather than IOMeter. Why? To explain our choice of the participating drives. The fact is the good old WinBench revealed a kind of anomaly that made us test more drives and even use more Serial ATA controllers! To be exact, we got unexpected results, rechecked and verified them for long, but didn’t find our mistake: the Serial ATA models had become faster! But let’s talk about that at the end of this section, after the analysis of the results.
Starting from this review, we will be doing our WinBench tests not on the entire capacity of the drive but rather on a single 32GB partition to put all the tested drives under the same conditions. The problem is that the size of the MFT or FAT is directly proportional to the capacity of the disc and becomes too big at such huge partitions as with the drives we are testing today (370GB!), leading to an undesirable growth of the overhead and measurement error range.
WinBench 99 results (click here to check out the table)
This review also differs from the earlier reviews of hard drives you’ve read on our site in the look of the diagrams. And I ask for your opinion if we should use such diagrams in our future reviews, too. The point of our innovation is putting both NTFS and FAT32 data into a single column of the diagram. Thus, the total number of pictures is reduced in double, without worsening their look (at least, they don’t look any worse for me).
Most WinBench subtests are highly sensitive to the capacity of the tested hard disk. The bigger the disc, the better results it achieves. That’s why I threw a HDS722516VLAT80 model into the fight. It is as many times smaller relative to the HDS722525VLAT80 as this latter is smaller relative to the Deskstar 7K400. Business Winstone didn’t become an exception this time around: the results of the 400GB model by far surpass the numbers of the previous models in both file systems.
The change of the controller leads to a sudden increase of performance in NTFS, but also to a small performance hit in FAT32. It’s normal since Business Winmark has always been fastidious about the controller, or rather about the controller’s driver. The different drivers account for the difference between the ATA and Serial ATA models in this case, as the results of the ATA model on the Promise SATA150 TX2plus, a nominally Serial ATA controller, prove (this controller has one classic ATA channel).
You can also see that the Serial ATA 7K400 model is slower than its ATA mate. This shouldn’t disturb us at all because the drive has to communicate with the system through an interpreter – the serializer chip.
The priorities of the Advanced Visualization Studio test are the drive’s linear speed and effective look-ahead reading. It is here that we see first signs of the revelation. The first UltraATA/133 model from Hitachi doesn’t work with the ATA/133 Promise Ultra133 TX2 controller confidently, while the Serial ATA version outperforms the ATA analog! The gap is small and is only observed in NTFS, but that’s only the beginning.
To make sure the controllers are blameless here, and we do see a difference between the generations of hard drives, we add the results of the Deskstar 7K400 and 7K250 on the ICH5 and on the Promise SATA150 TX2 controller, respectively. The latter pair isn’t impressive whereas the model of the Serial ATA controller doesn’t affect the performance of the Deskstar 7K400 much.
This time FrontPage produced a surprisingly clear picture that seems to need no comments at all (the reduction of the partition size to 32 gigabytes must have helped here). The only thing you may want to mark is the sudden failure of the Serial ATA controller integrated into the ICH5.
There are no sensations in MicroStation: the Promise SATA150 TX2 and the chipset’s Serial ATA controller are much faster than the Promise Ultra133 TX2 in NTFS. But this is only true for the Deskstar 7K400. The Serial ATA controllers give no advantages to the previous drive model.
Photoshop appreciates the physical qualities of the drive, its linear and data-access speed. Since these parameters have remained the same, there’s a very small gap between the controllers and the Deskstar generations. The 160GB model is evidently slower than the others, and it’s not quite clear why.
The Deskstar 7K400 suddenly accelerated in this video-editing program and left the 7K250 behind in FAT even on the Promise Ultra133 TX2 controller. Then the gap became even wider on the Serial ATA controllers. The controller’s influence on the result is obvious in this test.
The great lover of effective deferred write, Sound Forge puts the competing drives rather far from each other. And although the HDS724040KLAT80 wins greatly from switching to the Promise SATA150 TX2 controller, its Serial ATA analog (HDS724040KLSA80) proved to be even faster, and quite considerably so! The difference between the two is only in the interface speed (133MB/s with the ATA version and 150MB/s with the Serial ATA version) and the firmware algorithms – this fact needs some thinking over.
Compiling projects in Visual C involves reading/writing numerous small-size files, and we see the most impressive difference in the results here. It is over 50% between the Promise controllers in NTFS and is visible in FAT32, too. We haven’t yet seen such a thing in our labs. And the Deskstar 7K250 wins nothing from switching to another controller.
Now it’s time to do some local summarizing, especially since we’ve got such sensational results. According to the High-End Disk Winmark showing, the 7K400 seems to be as fast as the 7K250, but becomes much faster on a Serial ATA controller, even the Parallel ATA model! It wouldn’t be that weird (controllers are always different, you know), if it were not for some additional details, not easily discernable at first:
Solving this puzzle I can only come to the conclusion that the electronics with support of UltraDMA/133 is the underlying reason for all that. And really, the Deskstar 7K400 is no faster than the Deskstar 7K250 on the Promise Ultra133 TX2 controller, despite the advantage in capacity that directly affects the results of WinBench. But attached to the Promise SATA150 TX2 or to the ICH5, the Deskstar 7K400 is much better than the 7K250 in performance than it is in the capacity. Besides that, the Serial ATA versions are sometimes noticeably faster than the ATA versions where look-ahead reading or deferred writing is important. So, Hitachi made some serious changes in the electronics, that’s certain.
I’m going to start the IOMeter tests by exploring the fundamental characteristics of the drives. Creating a queue of sequential requests to read and write ever bigger data blocks, we can evaluate the efficiency of the interaction with the HDD electronics.
Sequential Read/Write results (click here to view the results table)
The reading performance of the 7K250 and 7K400 is so similar that the graphs merge in one. You may remember from our previous report that there were no differences between ordinary and adaptive look-ahead reading algorithms at linear reading, so the results shown in the diagram are quite comprehensible. But it’s anyway good that the new generation of electronics is no less efficient than the older one – we have often seen that new drive models would be slower than their predecessors in reading small-size blocks.
There are no big differences between the participating models at linear writing, either. The 7K400 is somewhat slower than the previous model, even though the gap is negligible.
The average read access time of the 7K400 hasn’t got any worse even though it uses five platters now. The HDS722525VLAT80 wins this test due to its “shortened” platters – its capacity is truncated to 250 billions of bytes for some reason, quite unusually for products from IBM/Hitachi.
But there are differences in the average write access time test. First, my point about the bad compatibility with the Promise Ultra133 TX2 controller is confirmed: the ATA version of the 7K400 is the slowest on this controller, while it performs quite well on the Promise SATA150 TX2. The Serial ATA 7K400 models, however, boast the highest performance. They are accompanied with the senior model of the 7K250 series which has the “shortened platter” advantage.
Next I calculate the ratio between read and write average access times. The result is indicative of the efficiency of sorting of deferred write requests.
The failure of the 7K400 on the Promise Ultra133 TX2 controller is already clear, while the fact that the results of the Serial ATA 7K400 and of the senior 7K250 model coincide means that Hitachi has really developed firmware with new dynamic look-ahead read and deferred write algorithms, but uses them in the flagship models only. The new algorithms should be most effective at processing mixed streams of requests. Let’s check it out right now in the Database pattern.
Database results (click here to view the results table)
As usual, I’m going to check the dependence of the performance on the number of write operations.
The drives differ little under the linear load, so I had to “zoom in” by using a not-from-zero scale. The graphs of all the models have the same shape, which is another indication of the close affinity between the Deskstar 7K400 and 7K250.
The senior 7K250 model is the first at the start, enjoying its advantage in average seek time, and the rest of the drives follow it in a dense group. Starting from about 40% writes, the 7K400 on the Serial ATA controller integrated into the ICH5 tries to fight for the first place, but the better average seek time says its word at 100% writes. At about the same spot the Deskstar 7K400 on the Promise Ultra133 TX2 controller starts to have problems. It first loses its advantage over the two-platter 7K250 and then falls behind the others at the end of the race.
No, let’s examine the scalability in load at various types of requests.
It’s impossible to defer read operations, so the average seek time determines the winner here. None of the drives has any problems with scalability at reading. The 7K400 models are all slightly slower than the 7K250 at a load of 256 requests, but this is not important at all.
There are differences in the mixed modes (the diagram shows averaged results for all write-to-read ratios). The adaptive algorithms win here, although the ATA version of the Deskstar 7K400 on the Promise S150TX2 controller is no slower at small loads.
The biggest gap, as might have been expected by the results of the earlier tests, occurs at 100% writes: the 7K400 on the Promise Ultra133 TX2 is hopelessly slow. It also has some problems with scalability in load on other controllers, too.
Well, now we can make some statements concerning the features of the Deskstar 7K400 electronics. Without doubt, the firmware is directly inherited from the 7K250. The introduction of the UltraDMA/133 mode doesn’t lead to any breakthroughs, but causes certain incompatibility with some ATA controllers at deferred write operations. It’s hard to name the real cause of this behavior but it’s not the first time the Promise Ultra133 TX2 controller acts up. We saw oddities even with the D740X, the first drive to support the UDMA6 protocol, so we can ascribe the blame for the strange results of the ATA version of the Deskstar 7K400 to the controller’s peculiarities. Curiously, the Deskstar 7K250, on the contrary, had problems with deferred writing on Serial ATA controllers (see our article called Hitachi Deskstar 7K250: Vancouver 3 HDD Review for details).
Now, I’m going to try the drives in patterns that emulate a complex load on workstations and servers.
Workstation, File Server, Web Server results (click here to view the results table)
Our Workstation pattern emulates a workstation under a heavy load of several applications running in parallel. Since the percentage of write operations is high here, the 7K400 on the Promise Ultra133 TX2 controller is expectedly slow. The other variants are equally fast, while the HDS722525VLAT80 wins through its advantage in the average seek time on “shortened” platters.
It’s all roughly the same in the File Server pattern despite the larger range of loads. Thanks to a smaller percentage of writes, it is the 160GB model that’s on the losing side now. So, we can talk about some progress of the Deskstar 7K400 over the 7K250.
There are no write operations at all on a web-server without active content (like conferences or other), so all the 7K400 come to the finish the same moment, but lose to both 7K250 due to a bigger average seek time. Well, it’s just more difficult to wield an actuator with ten read/write heads than one with four or six heads. But the difference is rather negligible, anyway.
To make the picture compete, I’m going to test the drives at real work with files. The updated FC-Test makes it an easy thing, thanks to its automatic reboots and more precise speed measurements.
The diagrams show not the absolute speeds but the arithmetic mean of the maximum results across all the patterns. This helps to get a more accurate view of the performance of the drive due to averaging possible fluctuations caused by random factors.
The tests of writing files of varying size prove the superiority of the Serial ATA Deskstar 7K400 models in NTFS. They are only slower than their mate attached to the Promise Ultra133 TX2 in FAT32, and thus become overall leaders. It means the Hitachi programmers have managed to improve the firmware because the Serial ATA versions of the Deskstar 7K250 don’t have any advantages over their Parallel ATA mates. In fact, we could have expected some gains from the higher interface bandwidth at writing in the first place: the data are being quickly pumped into the drive’s buffer and the system can get to its own business.
The Promise Ultra133 TX2 seems to have no problems at write operations with real files. But if we take a closer look at the results of separate patterns, we can see a curious picture: this controller is an unrivalled leader with the Deskstar 7K400 in FAT32, but it loses its speed abruptly on small files (Programs and Windows patterns) in NTFS. It’s almost the same with the Deskstar 7K250 while the Serial ATA controllers don’t slow down on smaller files. Thus, we can localize the problem with even more accuracy: the Promise Ultra133 TX2 takes too long to initialize a data transfer to the drive but then performs the transfer quickly. Thus, the efficiency of this controller directly depends on the amount of data to be transferred.
Sequential reading is a simple task, so the results please the eye of the tester with uniformity: the Deskstar 7K400 is slightly faster than its predecessor in NTFS, especially coupled with Serial ATA controllers. All participants are peers in FAT32, if we make allowances for measurement errors.
Copying files is a more complex task, with certain nuances possible. For example, I can remember hard drives from Samsung easily beat their opponents with a higher “linear” speed. Let’s see if there are any changes in this respect with Hitachi.
The changes are obvious even at copying from one partition of the disk to another: the Deskstar 7K400 on Serial ATA controllers are much faster in NTFS and remain overall winners, even despite the minor loss in FAT32. Note also that the Deskstar 7K250 are fastest in FAT32.
The Serial ATA controller from the ICH5 South Bridge lost its ground when copying from one partition to another. The Deskstar 7K250 almost overtakes it even in NTFS. This drive becomes an overall leader thanks to its excellent performance in FAT32.
So what is my opinion about the Deskstar 7K400? Yes, this is a narrowly specialized product since 400GB capacity is not in high demand everywhere, yet this drive has performed excellently across all of my tests.
Compared with the previous, Deskstar 7K250 model, the new product has become much faster in NTFS but somewhat slower in FAT32. This change of priorities seems justifiable considering that FAT32 is losing its popularity (for example, you cannot create a FAT32 partition larger than 32 gigabytes in Windows XP). What’s strange is the abandonment of adaptive algorithms in the firmware of the ATA version of the Deskstar 7K400. Maybe Hitachi is implying that it’s Serial ATA era already and pushing the users into it?
The changes in the electronics provoked by the introduction of the UltraDMA/133 data-transfer mode affect the interaction with the controllers quite unexpectedly. Where the Serial ATA Deskstar 7K250 used to be slower than its ATA version, the Deskstar 7K400, on the contrary, speeds up. The gain amounts to 60% at most. The developers’ work is truly astonishing, even though it’s not perfect: the performance has degenerated on some controllers. Well, the Deskstar 7K250 didn’t show its best on some controllers, either.
The combination of excellent performance with a number of exciting features like hot-swapping Serial ATA, active vibration compensation, and commands for processing streaming data would make the Deskstar 7K400 a versatile hard disk drive if it were not for its crazy capacity of 400 gigabytes! Let’s better wait for the updated Deskstar T7K250 with the new electronics?