by Alexey Volkov , Nikita Nikolaichev
07/19/2004 | 04:07 PM
The third installment of our investigation of speed characteristics of same-size hard disk drives reached the point of 160GB storage capacity.
<%BANNER[article]%>Our roundups of 80GB and 120GB devices may have already helped you in your shopping for a drive to use in your own system and this review is of less interest to you but still, although we have the same gang of manufacturers and well-known drive models, this 160GB HDD roundup is going to be different in several respects.
First, we test and compare 160GB devices, and this means you won’t see old models anymore, and in its turn it means we can expect a tougher competition here. Second, we start investigating devices of the same model, but with different firmware versions. This is going to spice up the review, that’s sure. Third, to hit the magic number “21” we added two drives of 180GB capacity into the show (we only found two 180GB devices for our test lab, so there was no sense in picking them out into a separate review).
Regardless of the impressive participation, we should make a reservation that this roundup is far from being a “full and comprehensive review of all possible drives from any possible point of view”. It is not because we are overmodest but rather that we are perfectly aware that our tests are only indicative of the speed characteristics of the drives, and only in the benchmarks we use.
We don’t measure the temperature of hard disk drives. Alas, our attempts to find any logic in those temperatures the drives report through SMART were a fiasco. Yes, all drives now report their temperature, but the problem is that the manufacturers put the thermal diode in different locations. Thus, the reported temperature differs considerably between manufacturers and cannot of course be used to compare the thermal characteristics of different drives. Right now we’re returning to the traditional and simple method of measuring the temperature of HDDs with an infrared thermometer, but time is required to gather some statistics.
We don’t also measure the noise the drives produce at work. There’s no sense in doing this with a simple audio-noise meter in a system case with a Pentium 4 (and an appropriate cooler) inside, while a good methodology of measuring noise requires both special equipment and special knowledge, which we don’t have right now.
If you’re still with us after so long an introduction, let’s muster the participants of our today’s tests.
Hitachi is represented in this review with three drives of the Deskstar 7K250 series and one of the Deskstar 180GXP series.
Hitachi doesn’t produce 5400rpm HDDs for desktop computers anymore, and its “value” models just have a smaller cache buffer. One “value” HDD from Hitachi takes part in our tests. The other three drives, including the IC35L180AVV207-1, have 8MB of cache memory. The basic distinction between the 180GXP and 7K250 families is in the capacity of the platters: the drive from the 180GXP series has three 60GB platters, while the other series uses 80GB platters.
Today, Maxtor is shipping three desktop models of 160GB capacity.
These drives all belong to the same family, but one of them, the 6Y160L0, has a smaller cache and 60GB platters (our method of determining the platter capacity of Maxtor drives is described here) The two remaining drives have a 8MB buffer and 80GB platters – good assets to start with.
Unfortunately, we couldn’t get the 5400rpm 160GB drive, the SV1604N model, into our roundup. So the company is only represented with three 7200rpm devices.
It should be mentioned that 160GB drives from Samsung, unlike those from Maxtor, cannot use three platters due to design considerations. So, we can be sure that 160GB HDDs from Samsung use 80GB platters.
Seagate has the widest presence in this review – we’ve got seven participants from this company but also one reservation.
Yes, there are only three drives, but we had two or three samples of each model with different firmware versions. We tested all the drives that we had on our hands to find that the firmware version affects the performance dramatically, as you will see later on.
All the drives are configured identically: two 80GB platters and four heads. The models differ in the amount of cache memory and the interface.
Western Digital is also widely represented, nearly like Seagate:
The drives differ in their design. Note that Western Digital, like Hitachi, produces a 180GB model (three 60GB platters). Then, the two drives with a small cache buffer differ between themselves: the WD1600BB uses 60GB platters, and the WD1600LB – 80GB platters. The latter model also features fluid dynamic bearings that reduce the noise. We have also got two “modifications” of the WD1600JD model – on 60GB and 80GB platters! To differentiate between drives on different platters we refer to them as WD1600JD/60 and WD1600JD/80, respectively.
We had to use two controllers as we had hard disk drives that connect across two different interfaces. So we took two controllers from the same manufacturer:
The testbed was configured as follows:
The drives had the following firmware versions:
We wrote the Maxtor drives through before the tests to avoid their forced write verification.
For WinBench tests we formatted the drives in FAT32 and NTFS as one partition with the default cluster size (FAT32 formatting was performed with Paragon Partition Manager). We ran the tests seven times each, chalking up the best result. The HDDs didn’t cool down between the tests. For the FC Test we partitioned the drive into two logical volumes, 32GB each. For IOMeter tests, we used Sequential Read, Sequential Write, Database, Workstation, File Server and Web Server patterns. You can refer to our previous reviews for details about the patterns.
The Database is traditionally the first pattern in our test program. It helps us to evaluate a drive’s performance with a mixed stream of requests for reading/writing random-address 8KB data blocks. By changing the ratio of reads to writes we can check the drive’s ability to sort the requests out.
The results table is very big, so we divide the results of this pattern in groups, by manufacturer.
For better readability, we draw diagrams for different workloads (the number of requests in the queue).
You can see it in the above diagram that the drive with a 2MB cache buffer starts losing to the other models from 50% writes on. It comes as all HDDs from Hitachi are very unwilling to give the cache segments reserved for reads, for writes. A drive with a small cache buffer has overall fewer cache segments than an 8MB-buffered device. When a drive receives only write requests, the drive’s firmware uses the whole cache to store writes, and the random write speed increases suddenly, due to the more efficient work of deferred write algorithms.
The IC35L180AVV207-1 model from the older Deskstar 180GXP series loses to the members of the new Deskstar 7K250 family when there are many writes to be performed. As for a difference between the 7K250 disks with an 8MB buffer but different interfaces, there is none.
We increase now the load to 16 requests:
The drives all have the same speed in the random read mode, but split up in two groups as soon as there appear write requests. The first group is comprised of the 7K250 disks with an 8MB buffer – their speeds are actually much alike.
The remaining two, the IC35L180AVV207-1 and the HDS722516VLAT20, started out together, but the latter couldn’t keep the pace on due to its small cache. You should also note that the HDS722516VLAT20 suddenly accelerates at the 100% writes point, just it did under the 1-request load, and stands even with the other drives there.
The workload being the highest, the graphs practically remained the same as under the 16-requests load.
The same Database pattern as performed by the Maxtor team:
We construct some graphs:
The victory of the 6Y160L0 disk over its potentially mightier mates can only be explained by the fact that the three-platter 6Y160L0 uses short platters, with fewer tracks per each. Thus, this drive has a smaller access time compared to the other Maxtors, and this parameter helps it win in this test.
We make the load a bit heavier:
The leader remained the same, and the other two HDDs exchange their places as they transition from pure reading to pure writing – they can’t catch up with their junior brother again.
The 6Y160L0 is a bit slower at pure reading than the other two, but regains its leader position as soon as there are writes in the queue. The 6Y160P and the 6Y160M0 worked practically identically through this test. We should also note that the Maxtor drives have smaller maximum speeds than the Hitachi team.
Let’s discuss the Samsung gang now.
They perform like a team of synchronized swimmers. Irrespective of the buffer size and interface, they score much the same results.
It’s roughly the same under the 16-requests load; only the SATA drive is somewhat faster at reading and slower at writing.
There’re no big changes at the maximum load, either. The SP1604N has a small advantage working with a mixed request stream, but joined the rest of the group at 100% reads and 100% writes.
It’s time now for the Seagate crew to enter the scene:
It’s hard to read such tables – we prefer watching diagrams:
The graphs are indicative of the dependence of the performance on the firmware version. Seagate’s SATA drives are evidently faster than their PATA analogs. Well, we already discussed this fact in a separate review, so it’s more interesting to clear out the situation with different firmware versions.
For example, the ST3160023AS model showed its highest speed in this test with firmware 3.05, and, accidentally, it was the fastest of all the Seagate drives! Curiously, the later versions of the firmware (3.14 and 3.18) fell behind the leader in high-percentage-of-writes modes as well as in the random read mode.
As for the PATA devices, the ST3160021A and the ST3160023A with firmware 3.06 perform much alike. Probably, the same firmware has a greater effect on the performance than the amount of cache memory on board.
We increase the load…
The drives remained at their positions, only the gaps have grown somewhat. It’s clear that the ST3160021A works a little slower with firmware 3.04 than with firmware 3.06, while the ST3160023A with firmware 3.71 has a strong dislike toward write operations.
All the drives did equally well in the random read mode here but the SATA-interfaced disk is ahead as soon as there appear write requests in the queue. We can put it in another way: “the PATA disks slow down on write operations”.
Western Digital is the last manufacturer on our list.
The request queue contains only one request at a time:
Unlike hard disk drives from other makers, the WD team forms a dense group. The WD1600JB rises above the crowd somewhat, and the WD1600JD/60 worked well at high writes percentages. On the contrary, the WD1600JD/80 is rather poor at writing.
The positions of the drives haven’t changed under this higher load, only their read speeds increased from 75 to 100 IOps. Now we push the load up to 256 requests:
The speeds of the drives grew up and that’s about all to that. We can only single out the 1600LB that worked well in modes with a high writes percentage.
The hard disk drive is receiving a stream of read/write requests in this test, the request queue always remaining four requests long. Once every minute, the size of the processed data block changes, so we get the correlation between the linear read (or write) speed and the data chunk size.
Click to enlarge
The Hitachi IC35L180AVV207-1 is reading small data blocks (up to 4KB) better than the others, but then yields its leadership to the HDS722516VLAT20 model from the same company. The Maxtor 6Y160P0 and the Samsung SP1614C did the reading quickly enough, too.
Irrespective of the firmware version, the Seagate ST3160023AS works badly with small-size blocks, while the WD 1600JD/60 is the slowest on large blocks – take note of this curious fact.
Once again we split the participants up into groups by manufacturer for better readability of the results.
You can view the graphs by the following links:
Let’s see how these results change at writing:
Click to enlarge
The Maxtor 6Y160P0 is the fastest at writing data blocks of any size but the three disks from Samsung are following it closely.
The Western Digital 1600BB and 1600LB handle small blocks very badly, and the Seagate ST3160023A with firmware 3.71 is disgustingly slow at writing big data blocks. In the diagram below you may see that this disk cannot exceed a barrier of 20MB/s. Other disks from Seagate don’t show this behavior, so we are inclined to think that the firmware is to bear the blame. Curiously, one and the same firmware version 3.71 acted up both at random requests (Database pattern) and at sequential ones…
The diagrams follow:
The next patterns simulate the typical load on the disk subsystem of a server.
We compare the drives by averaging their speeds under four workloads. This performance rating doesn’t include the 256-requests load.
Four drives on top have left the rest of the participants far behind. They are: two Hitachi HDDs, the Seagate ST3160023AS (with firmware 3.05) and the WD1600JB.
The three samples of the ST3160023AS model from Seagate with different firmware versions allow us evaluating the influence of the firmware on the performance in this pattern. Obviously version 3.05 handles this pattern better than the others – the drive with firmware 3.18 found itself six steps below it in the results table. The third ST3160023AS, with firmware 3.14, is three positions lower yet.
Next goes the Web Server pattern:
That’s a curious outcome: the Hitachi team occupies the pinnacle of the podium, and the Seagate ST3160023AS in its three incarnations has nestled a little lower. The bottom of the diagram mostly consists of the remaining PATA drives from Seagate and two devices from Maxtor.
This pattern loads the hard disk drive with a lot of write requests, so the drive’s firmware must do deferred writing well in order to succeed here.
We see that firmware 3.71 of the Seagate ST3160023A model is poor compared to the other versions. Write requests slow this model down dramatically. Let’s take a look at the performance ratings:
The WD 1600JB drive is the best here, yet we might have guessed it by the results of the Database pattern. The Maxtor 6Y160L0 is a mild surprise – the firmware of this drive is making a most efficient use of its small 2MB cache buffer. Strangely enough, the best of the Hitachi team took only the 11-th place – they seem to dislike mixed modes (with both reading and writing). The Seagate ST3160023AS with firmware 3.05 holds the second position.
Now we reduce the address space of the drives to 32GB and repeat the test.
The Seagate ST3160023A with firmware 3.71 loses hopelessly here, even to the other PATA drives from the same company.
It seems like we have been too hasty dismissing the model with firmware 3.18 – it did quite well in the restricted 32GB space. As usual, the Samsung disks are good, especially the SP1614 that took the second place. The “old” Hitachi IC35180AVV207-1 drive is the third (don’t forget it features a three-platter design, which reduces its access time).
This was the last of synthetic IOMeter patterns; we can now proceed to WinBench 99.
We use WinBench to check the hard disk drives in the “desktop PC” mode. We format the disk in the NTFS file system by means of the system tools (the default cluster size is 4KB) and in FAT32 using Paragon Partition Manager (the cluster size is 32KB). We also perform our tests on 32GB capacity in NTFS and FAT32 file systems (partitioning the drives be means of the standard Windows 2000 Disk Manager).
We first discuss the physical parameters of the HDDs, which don’t depend on the file system.
The Seagate ST3160023AS (firmware 3.05) has the best average access time, and this explains its success in server-simulating patterns. The three-platter devices – Maxtor 6Y160L0 and Hitachi IC35L180AVV207-1 – have a slightly higher average access time. Then, the Hitachi team comes in a dense group – they traditionally boast a good access time.
The PATA drives from Seagate are in the bottom of the diagram: the speed of moving the heads around has been sacrificed for their noiselessness…
The results in this test directly depend on the areal density per platter, so only modern drives with 80GB platters can hope for a victory here.
That’s right – the top of the diagram is besieged by high-areal-density drives, whose results differ but slightly among themselves. Three-platter devices gathered up at the bottom of the table, mingled with the Seagate team. The poor results of the drives from that firm can easily be explained if you take a look at their data-transfer graphs.
HDDs with “shortened” platters can be easily spotted in the diagram. They don’t use the full capacity of their platters and have a very high read speed at the end of the disk! Clearly, the WD1600JB and Maxtor 6Y160L0 fall into this category – both drives have three platters and six read/write heads, so the capacity of their platters is not even 60GB, but rather 53GB.
Note also the speed at which the Seagate ST3160023AS with firmware 3.05 is reading the last tracks – considering this abnormally high speed, we can say it is a cut-down drive, too! Is it a three-platter device, too?
But that’s impossible, since Seagate Barracuda 7200.7 disks use two platters (at least, so far)!
It turned out simple: if you examine the data-transfer graph of the Seagate ST3160023AS with firmware 3.05, you may notice that the disk hadn’t been recognized for its full capacity, 160GB! It’s incredible, that it is a fact, nonetheless. Until now, we thought the Promise S150 TX2 Plus controller had no problems identifying a disk’s full capacity – that’s exactly why we used this controller in all our tests.
An update of the controller’s BIOS may solve this problem, but it’s also possible that such an update would affect the controller’s own speed. If it does, would we be right to compare the old results (with the old controller’s BIOS) with new ones? We decide to keep the old BIOS for the time being, as this problem only occurred with one hard disk drive and with one definite firmware version.
Getting back to the Seagate ST160023AS model with firmware 3.05, we had to repudiate the results of that drive. They were overestimated because the disk’s address space was not used to the full – the farthest tracks were “cut off”, resulting in a lower average access time. Just take a look at the Disk Access Time diagram to see how much the ST3160023AS with firmware 3.05 got from this incorrect capacity detection – 0.7msec. It seems like a trifle, but this drive had a big advantage over the others in tests that were tightly tied to the average access time parameter.
You can view the data-transfer graphs we got with the help of WinBench 99 by the following links:
The following tests don’t care about the “size” of the drive (the difference between a 160GB and a 137GB drive is unimportant), so we can take the Seagate ST3160023AS fw 3.05 back into our competition.
Let’s examine the results in the FAT32 file system. We got a mountain of numbers, so we divided the results table into parts and linked to them:
It took a while to put all the data into diagrams, but did it anyway. Let’s now discuss the drives’ results in two integral tests: Business Disk Winmark and High-End Disk Winmark.
Maxtor’s 6Y160M0 and 6Y160P0 took the first two positions in the High-End test, far ahead of the rest of the drives. The Samsung SP1614N climbed its deserved third place, with a good gap from the others, too.
Western Digital got the podium in the Business test with its WD1600JB, WD1600JD/80 and WD1800JB models.
Now, let’s run these tests again in NTFS.
You can see the full data by the following links:
Maxtor’s 6Y160M0 and 6Y160P0 not only kept their leadership, but also widened the gap. The Samsung SP1614N lost its third place to the Hitachi IC35180AVV207-1. In the Business test, Western Digital couldn’t confirm its superiority – its best disk, the WD1800JB, occupies the fifth place only, while the first position is shared between the Maxtor 6Y160P0 and the Hitachi HDS722516VLAT80.
We decrease the disk space to 32GB and run the tests once again.
Save for minor changes, the positions remained the same; only the gaps between the drives diminished somewhat. The two Maxtors remained in the lead, while the SATA-interfaced disk from Samsung, the SP1614C, stepped up to the third place.
Let’s repeat the tests on a 32GB partition in NTFS:
Like in the previous case, there’re no surprises – the top of the diagram is still the domain of Maxtor, and the Samsung SP1614N again stepped once position higher. Otherwise, it’s all the same…
For detailed results of WinBench 99 you can follow this link.
Now we’ve approached our most exciting File-Copy Test.
This is going to be our last article to use the FC Test version 0.5.3 – all subsequent reviews will offer you the results of the FC Test version 1.0. We mention this fact to confess to our earlier mistakes: read speeds are somewhat overestimated in this review since we didn’t restart the computer between file creation and reading operations.
Anyway, we wield this testing tool according to our traditional methodology: we create two logical volumes, 32GB each, and format them in NTFS and FAT32. Then we create a set of files on the first logical volume, and read it from the drive, and copy it into a folder on the first logical drive (within one partition – copy near) and on the second logical drive (to another partition – copy far). We perform these operations with five file sets:
We start out with the NTFS file system, much popular today, thanks to Microsoft. We’ve got quite a lot of drives, so we’ll be discussing each operation for each file set independently. So, we first create a set of files on the disk.
The creation of the Install pattern files:
The drives from Maxtor are the best at creating Install files. Only two devices from Samsung (SP1614C, SP1614N) and one Hitachi (IC35L180AVV207-1) try to show some competition. That’s not a surprise – we saw these models getting to top positions in previous tests, too.
Next goes the ISO pattern:
The two Maxtors have an even greater advantage at writing ISO-like files. The Samsung SP1614N outperforms the Hitachi IC35L180AVV207-1. Something is wrong with the rest of the Hitachi team, the members of the Deskstar 7K250 family – writing big files doesn’t seem to be among of their fortes.
The six leaders remain at their positions creating MP3 files, while the seventh-place Seagate ST3160023AS with firmware 3.05 (remember that tricky drive?) is closely followed now by the other drives of that model but with different firmware versions as well as by the ST3160023A-fw3.06. So we’ve got a Seagate clan right in the middle of the diagram.
When creating small files of the Programs pattern, the Hitachi team woke up. The IC35L180AVV207-1 rose to the third position, and the other three models made it to the “top ten”. The WD1600LB suddenly improves its standing, soaring up from the 18-th to the seventh place.
The rather old Hitachi IC35L180AVV207-1 drive speeds up in the Windows pattern to step on the very top (we remind you that this drive has three platters and thus has an advantage over two-platter models). The drives from Samsung missed that spurt and even let the three-platter Maxtor 6Y160L0 to come ahead, in spite of its small 2MB buffer.
This was the last file creation test; we’ll be doing some reading now.
Let’s examine the results pattern by pattern:
The Hitachi HDS722516VLSA80 and the Maxtor 6Y160M0 read the Install pattern fast, but the same drives with the PATA interface somehow fell behind. The Seagate ST3160021A (with firmware 3.04) took the third place, while another version of the firmware allowed this drive to take 11-th place only.
Two devices from Samsung took the top of the podium, while the ex-leader, the SATA-interfaced
Two teams – Hitachi and Maxtor – are again victorious in the MP3 files reading test. The failure of the Samsung drives is quite inexplicable – the best of them, the SP1614N, is only the seventh, while the same drive but with the SATA interface is the 17-th!
The Hitachi HDS722516VLAT80 has a perceptible advantage over the rest of the drives at reading the Programs pattern. The Western Digital team has a meeting at the bottom of the diagram – they have a hard time trying to read small files.
The drives from Maxtor and Hitachi are contending for the leadership in the Windows pattern. Samsung’s and Seagate’s devices are in the middle of the diagram and the Western Digital team is again down at the bottom.
That’s enough for reading; let’s get to copying files within one partition.
No disk could get twice on top, and there are even two winners in the Programs pattern, so we’ve got some violent competition here.
The Install pattern comes first, as usual:
The drives from Samsung, which were in the middle of the diagram at reading, are now settled at the top two places, although closely followed by their rivals.
Hitachi drives’ dislike towards large files showed itself at copying ISO-type files. That’s why there are two disks from Seagate in the top part of the diagram. The leading disks from Samsung exchanged their places. The Western Digital team started regaining positions from the Seagate team – how long are they going to do that?
The leader has changed. The pair of Maxtor drives outperformed the pair from Samsung. Interestingly, Maxtor’s SATA drive is faster at copying MP3 files, but it’s vice versa with Samsung: the PATA drive is faster than the SATA one.
As soon as it came to copying small files, the disks from Hitachi got moving and gained a tremendous speed, outperforming the other drives much. The Seagate ST3160023AS is surprisingly fast in the Windows pattern, and with any firmware version! Thus, we can say that this operation greatly depends on the average access time.
The last action of the FC Test is copying files to another partition.
We see the Samsung team dominating on large files and the Hitachi team on small ones. The drives from Seagate are poor again.
The drives split into pairs (PATA and SATA variants). The Samsung drives are on top, followed by the Hitachi devices; the two Maxtors are behind. Interestingly, it is only in the Samsung team that a PATA drive outperforms a SATA one.
When copying ISO-like files, the drives from Hitachi gave way to the Maxtors as well as to the Samsung SP1604N. The bottom of the diagram is mostly comprised of Seagate disks, with two Hitachi devices among them, but the latter fact is rather an exception to the rule.
The results of copying MP3 files practically repeated the results of the Install pattern, except that the pair of Maxtors is still faster than the Hitachi HDDs. MP3-like files are relatively big, and, as we’ve already learned, Hitachi’s HDDs don’t work well with such files.
No one can challenge the Hitachi team at copying small files (Programs and Windows patterns). The Seagate ST3160023AS with any firmware, although couldn’t reach the speed of the Hitachi drives, improved its results considerably and first caught up with the pair of Samsung devices and then left them behind.
That’s the end of our NTFS tests. The results for the FAT32 file system are similar to those we saw under NTFS and we don’t think it necessary to discuss them at length, just a couple of facts: first, the speeds are overall higher in FAT32 than in NTFS, and, second, the drives from Maxtor work better in NTFS than in FAT32, so the Hitachi team mainly competes with the Samsung devices in patterns than consist of small files.
It’s a daunting task to compare so many hard disk drives, as you can’t usually find a device that would be the best or near best in all the tests. So, we will try to name the winner in each of our benchmarks.
The drives from Hitachi looked well in IOMeter (File Server and Web Server patterns) – the drive of the older generation, the IC35L180AVV207-1, was especially good due to its small access time.
The Seagate ST3160023AS with firmware 3.05 did well in that test, too, but these high results were because of the unfortunate incident with the Promise controller (it didn’t recognize the full capacity of the drive). That’s why we exclude that drive with that firmware version from further discussion.
Thus, the drives from Hitachi will be the best choice for a server; the SATA drives from Seagate are the second best. On the contrary, the PATA drives from Seagate are most inefficient at server applications, and the Seagate ST3160023A with firmware 3.71 is the worst of them all.
Devices equipped with an 8MB cache buffer take the lead in WinBench 99. We’d like to especially mark the performance of the Maxtor drives that won High-End Disk Winmark and the WD team that showed a nice speed in Business Disk Winmark.
The Hitachi drives were only efficient on small files in the FC Test. The interface didn’t practically affect the performance there. The drives from Samsung are most stable under any load, irrespective of the file size. The Maxtors are somewhat unsure working with very small files and in the FAT32 file system.
So, among the reviewed devices, we recommend purchasing the Maxtor drives for work with streaming video/audio content, the Hitachi and Western Digital drives for work with common Windows applications. Samsung’s devices can be characterized as “universal” – they don’t seem to have any obvious weaknesses.