by Alexey Volkov , Nikita Nikolaichev
04/15/2004 | 01:43 PM
Every time we changed, however slightly, our testing methods or shifted to another testbed, we found ourselves where we had started – all old test results would become void, while new results are too scanty for an accurate comparison of different devices. Such a change of the testbed happened some time ago and again it told negatively on how fast the new reviews appeared as well as on the number and scope of hard disk drives we include into our tests.
Anyway, we worked hard to test everything possible and now we have accumulated tons of numbers for each category of hard disks. It’s time to juggle with them! :)
It was easy to choose the subject of the today’s roundup: every hardware forum has topics like “Which drive of xxx capacity is better?” That’s quite natural for users to ask for advice from more informed community members. On the other hand, few users have the opportunity of forming a personal opinion about all hard disk drives available in the market, so such topics usually degenerate into an argument of several people who are all claiming that their personal experience is the only “truth”.
Yes, we have more opportunities for reviewing and testing more devices than an average user has, but we’ll anyway try to refrain from inferior-superior judgments, but rather offer you the results of our tests with some comments explaining them.
This article is actually intended to be the first of a series of HDD roundups. Today we will compare as many as twenty hard disk drives of 80GB storage capacity, as this capacity seems to be the most popular for today.
The current market of hard disk drives for desktop computers consists of five manufacturers: IBM-Hitachi, Maxtor, Samsung, Seagate and Western Digital. We will have new models from each of these companies as well as not very new drives (I think meeting your own old HDD in this review will somewhat enliven this boring reading for you…)
Somehow we found ourselves having four HDDs from each of the manufacturers, so we are beyond any suspicions as to favoring someone of them. Meanwhile, the manufacturers are not all represented with equivalent (of the same generation) devices. Yes, it’s really hard to make a truly comprehensive roundup – we just couldn’t find a few interesting models in time for this test session.
The technical characteristics of the drives, declared by the manufacturers, are listed below, in brief:
A member of the DTLA family may look strange standing next to 7K250 drives – there are three generations in-between (60GXP, 120GXP and 180GXP). However, the DTLA (75GXP) drive made a kind of revolution on its day, so we will use it as a reference point to see the progress made in HDD making in the last years.
Besides the DTLA, we have an IC35L090AVV207-0 drive, a member of the 180GXP series, and two devices from the 7K250 series with ATA/100 and SATA interfaces.
The Maxtor team consists of three drives from the DiamondMax Plus 9 series plus the D740X-6L “oldie”. The last drive is strong enough, as you will see later!
Alas, we couldn’t get modern 80GB HDDs from Samsung with an 8MB cache buffer, so we don’t have the fastest Samsung products in this roundup. On the other hand, we have the results of Samsung’s drives with 5400 spindle rotation speed. I think it’s interesting to see how much slower they are against drives with a faster spindle.
Seagate is represented in full with its modern drives. Firstly, it is the only company that has two SATA drives in this test session, although they belong to different generations. Secondly, both SATA drives are equipped with an 8MB cache buffer. Let’s see how they perform in the tests.
We added a peculiar device to the three Caviar family drives from Western Digital – the WD740GD (Raptor). The distinguishing feature of the Raptor is its phenomenal speed – 10,000rpm! Of course, this drive is a potential winner of our today’s tests, but we are going to see soon that raw speed is not the only factor that matters. Note also that the Raptor produces more noise and vibration than 7200rpm hard disk drives.
We had to use two controllers as we had hard disk drives that connect across two different interfaces. So we chose a Promise Ultra133 TX2 and a Promise SATA150 TX2 plus.
The Promise Ultra133 had the BIOS version 2.20.0.14 and the driver version 2.0.0.29.
The Promise SATA150 had the BIOS version 1.00.033 and the driver version 1.0.0.27.
The testbed was configured as follows:
We used the following benchmarking software:
We wrote the Maxtor drives “through” to avoid the forced write checking (click here for details). The 6L080L4 drive also underwent special therapy – 10 “cold” cycles of start-stops for disabling write checking.
We formatted the drives in FAT32 and NTFS as one partition with the default cluster size (FAT32 formatting was performed by Paragon Partition Manager). We ran the tests seven times each, taking the best result for further analysis. The HDDs didn’t cool down between the tests. For FC-Test we partitioned the drive into two logical volumes, 32GB each. For Intel IOMeter tests, we used Sequential Read, Sequential Write, Database, Workstation, Fileserver and Webserver patterns. You can refer to our previous reviews for details about the patterns.
The Database test opens up the show. This pattern serves to reveal the ability of the drives to process a mixed stream of requests for reading and writing random-address 8KB data blocks. By changing the ratio of write and read requests we can estimate how well the drive is sorting the requests out.
For a more illustrative analysis, we draw diagrams for different workloads. The drives are grouped according to the manufacturer:
All drives showed similar speeds at 100% reading, but as soon as there appear write requests the IC35L090AVV207 model becomes the leader until 50% writes. After this mark, Hitachi HDS7225080VLSA80 proves that a large cache buffer and efficient lazy write algorithms may be more important for the performance than a smaller average access time (the IC35L090AVV207-0 has two platters and three read/write heads; considering the 60GB capacity of its platters, we can talk about it having small access time as the operational zone of its patters is narrower).
It’s interesting that Hitachi HDS722580VLAT20, which looks like the closest relative to HDS7225080VLSA80 model, behaves quite differently under this workload. It dawns upon this drive that it can save write requests in the buffer till the right moment only after the writes share equals 100%! Of course, we realize that the number of cache segments differs greatly between senior and junior models from IBM-Hitachi, just because the junior models have a smaller cache, 2MB.
Let’s increase the workload a little:
Under the workload of 16 requests, the SATA drive has hard time getting through the mixed read/write modes, losing even to the IBM DTLA307075. However, HDS722580VLSA80 restores its reputation in modes with more writes. Hitachi HDS722580VLAT20 was also slower than the DTLA307075 when there are more writes in the queue, but outperformed it subsequently. IC35L090AVV207-0 was the best, just a little behind HDS722580VLSA80 in two modes only.
Now, the workload is the highest:
The drives remained on their respective positions, only the gaps between the speed graphs have become wider. IBM IC35L090AVV207 draws a curious graph – this drive must have given us the utmost its mechanics allowed.
Now, let’s take a look at the Maxtor products.
Like with the IBM-Hitachi drives, we will go a little bit more into details.
Here, three drives out of four show similar results and it’s no surprise since they come from the same family, the DiamondMax Plus 9. The fourth drive behaves differently. Maxtor 6L080L4 was developed by the Quantum engineering team and their swan song was a success. This drive has the best seek time (11.7msec) of all 7200rpm IDE drives.
As you see, 6L080L4 drive is the fastest of all Maxtor devices in all modes, save for the random write mode. 6L080L4 doesn’t use lazy write algorithms, but processes all requests in the order they are received – this fact explains its bad results with 100% writes.
Under higher workload, DiamondMax Plus 9 with the SATA interface took the lead, although the “old” Maxtor (of the D740X series) was just a little behind it. Interestingly, the two remaining Maxtors from the DiamondMax Plus 9 series showed close results, in spite of their difference in the cache buffer size. At the same time, they were both far behind the 6Y080M0 model! This fact cannot be explained by our Maxtor 6Y080M0 having 60GB platters (i.e. it is has a “cut-down” platter, just like IBM IC35L090AVV207-0), as our 6Y080L0 also has 60GB platters – that’s obvious from the WinBench Disk Transfer Rate diagram…
A thrilling thing happened under the maximum load: Maxtor 6L080L4 is on top again, leaving all others behind. The old beat the young!
Now, let’s see how hard disk drives from Samsung handle this pattern.
It was hard to imagine just a couple of years ago that Samsung drives would be suitable for server systems. Times have changed and, unlike us, for the better. Samsung HDDs have been enjoying success recently. They are reliable, fast, cool, noiseless. Let’s watch them getting through this pattern.
You can see the Samsung drives progressing in speed with every new model. The members of the new generation, P80 and V80, are evidently faster than the drives from the P40 and V40 series. The shapes of the disk graphs are close to the ideal.
Seagate is the next company in our list.
And the diagrams:
Three drives, except ST380011A, form up a dense group under all workloads and it’s difficult to single out a leader among them. The worst model is evident, though. It is the ST380011A. Alas, this drive, like Hitachi HDS722580VLAT20, seems to have been “slowed” down artificially.
The last, but not least, manufacturer in our list is Western Digital.
There is a potential leader, WD740GD, among the company’s products. We are interested to see whether the new model, WD800LB (80GB platters, 2MB cache buffer, fluid bearings) can compete with the drive from Caviar SE series (WD800JB).
Yes, Raptor is far ahead of the rest of the drives under any workload. Three remaining drives match each other, although WD800LB turned to be faster than the oldies, notwithstanding a higher data density.
Intel IOMeter is sending a stream of read/write requests with a request queue of 4. Every minute, the data block size changes, so we get the dependence of the linear read (write) speed on the size of the data block.
Click to enlarge
The worst results are highlighted with red, the best results are highlighted with blue.
The table shows that WD740GD has the highest read speed on large data blocks, while Maxtor 6L080L4 did a good job with small blocks (up to 2KB). As for the worst results, Samsung SV8004H was the slowest on large data blocks (well, it is an old drive with 5400rpm rotational speed), while Seagate ST380013AS suddenly appeared the slowest on small data blocks.
Again, it’s not quite appropriate to put all results into one diagram, so we split the drives into groups according to their manufacturer. Here are the diagrams:
Now, let’s see how these results change in the Sequential Write pattern.
Click to enlarge
The leader remained the same – WD740GD is still the fastest at handling large data blocks. The Maxtor team has the best drive for dealing with small data blocks – 6Y080P0. The drives from WD displayed a strong dislike towards small data blocks, while IBM DTLA307075 suddenly fails on large blocks.
You can view the diagrams by clicking on the manufacturer name below:
Now we are going to check out patterns that simulate the work of the disk subsystem of a typical server.
We will compare the drives by calculating their performance ratings – the average of the drive speeds under four workloads. We decided to exclude the results under the 256 requests workload.
WD740GB goes unrivalled, but Maxtor 6L080L4 surprised us by taking the second place. The two 5400rpm drives from Samsung are closing the table, but Seagate ST380011A is not much faster than them…
Raptor HDD remains the leader in the Webserver pattern, followed by three IBM-Hitachi drives. Two relatively fresh products from Maxtor have got a surprisingly poor performance rating, while their older mate is still in the “top five” (due to its good access time, of course).
This pattern features a lot of write requests and this should mean some changes on the podium, although Raptor is very unlikely to be defeated by anyone else.
Yes, WD740GD is unreachable, but other drives changed their positions in the results table: IBM-Hitachi drives gave up the fight and didn’t even make it to the “top five”. Maxtor 6L080L4 continues to surprise us – its good results have become a tendency.
Now we reduce the address space of the hard disk drives to 32GB and repeat the test.
By narrowing the operational zone for the disks we boosted their performance. The ranks remained the same, only Maxtor 6Y080M0 dropped from the fifth to the eighth place.
This is the last synthetic Intel IOMeter pattern; we can now switch to WinBench 99.
We use the WinBench test to check out the hard disk drives in the “desktop PC” mode. We format the disk into the NTFS file system by means of the system tools (the default cluster size is 4GB) and into FAT32 using Paragon Partition Manager (the cluster size is 32KB). We also perform our tests on the 32GB storage space in NTFS and FAT32 file systems (partitioning the drives be standard Windows 2000 Disk Manager).
The table with benchmark results was very huge, so we split it into logical groups and linked to them:
FAT32:
FAT32-32GB:
NTFS:
NTFS-32GB:
It was very hard to combine the huge masses of numbers into diagrams, but we did it and are now ready to discuss the test results for FAT32.
The results of this test don’t depend on the disk storage capacity or file system, so let’s discuss them before the results for the FAT32 system.
As you see, there is no rival among 7200rpm HDDs to the Raptor drive in terms of linear reading speed. What’s characteristic, WD800LB drive was the second: Western Digital implemented 80GB platters into its devices later than the competitors, but the implementation seems perfect.
Going down the results table, we see 80GB-platter drives and devices with fast 60GB platters. All old HDDs and 5400rpm devices quite deservedly occupy the bottom of the table.
You can view the linear read graphs for each of the drives by clicking on the following links:
For example, the shape of the Maxtor 6Y080L0 graph explains why a drive with a true 80GB platter is so slow at linear reading in the beginning of the disk. We are just unlucky to have a bad specimen…:(
Let’s now compare the drives according to their average access time:
Again, WD740GD is nearly twice as fast as ST380011A (the junior drive from the Barracuda 7200.7 series). By the way, this drive showed higher access time than the Samsung drives with lower spindle rotation speed! In our Barracuda 7200.7 HDD Family Review we supposed that Seagate uses a special noiseless seek mode for its ATA drives. This reduces the noise from the device, but also trims down its performance.
Note also that SATA drives all took top positions in the results table. Well, fast seek has always been important for servers, even entry-level ones.
Let’s now consider the results in two integral tests: Business Disk Winmark and High-End Disk Winmark.
Our leader, WD740GD, feels some competition in the High-End Disk Winmark test – Maxtor 6Y080P0 is very close. The Maxtor drive is followed by the drive from the same family, but with the SATA interface, the 6Y080M0 model. However, the two of them are not very good in Business Disk Winmark (fourth and fifth positions, respectively). The second place (I guess you understand who the winner is) goes to WD800JB, while WD800LB doesn’t show remarkable results.
Evidently, WinBench favors drives with an 8MB cache buffer.
Now, the same tests are run on a logical volume 32GB big.
As the size of the disk drive is getting smaller, we don’t see any significant changes, only Maxtor 6Y080P0 and 6Y080M0 swap places in Business Disk Winmark test.
Let’s turn to the NTFS file system now.
Speeds have subsided in comparison to FAT32 and the gaps between the drives have diminished. Six out of seven HDDs with an 8MB cache buffer take top positions in the table. Strangely enough, Seagate ST380023AS is only the twelfth in High-End Disk Winmark, although takes the sixth place in Business Disk Winmark.
Concluding our WinBench 99 tests, we offer you the results of the drives in NTFS on a 32GB logical volume.
I think it is all clear without my comments. Let’s get to our File Copy Test.
That’s the most exciting part of our roundup. We run FC-Test according to our standard methodology: we created two logical volumes, 32GB each, on the HDDs and formatted them in NTFS and FAT32. We created a set of files on the first logical volume and read it, then copied it into a folder on the same volume (i.e. within one partition) and to another partition. We use five sets of files:
NTFS starts the tests.
Because we have 20 HDDs participating, we will discuss the results in detail pattern by pattern. So we start by creating a set of files on the drive:
WD740GD is the fastest at creating files from the Install pattern, although it doesn’t have its usual colossal advantage over the rest of the HDDs. Three Maxtors are closely following the Raptor! Let me remind you once again that the drive from Western Digital rotates its platters about 40% faster than its closest pursuers do. So the HDDs from Maxtor show simply astonishing results for their class.
Maxtor 6Y080P0 gets very close to WF740GD on large files like ISO ones. The gap is smaller than the measurement error for this test. In other words, we can say that these two drives show the same performance.
Finally, we see what seemed incredible. Maxtor 6Y080P0 is faster than the Raptor drive! I wonder if it is just a coincidence. Let’s see the drives processing smaller files.
Faster heads positioning allows Raptor to regain its leadership and increase the gap. However, the stubborn Maxtors are all following the leader. 6Y080P0 loses its position to the SATA mate, which has better access time.
We have the same group of leaders, only Samsung SP0802N with its scanty 2MB of cache memory suddenly takes the third place! I wonder what SP0812N and SP0812C would show us if they were participating in this test session?!
Let’s now see the drives reading the file sets.
Well, that’s the first surprise! Maxtor 6Y080P0 is on top, followed by two Hitachi HDDs from the Deskstar 7K250 series. Raptor arrives only the fourth.
This test is full of surprises. Raptor HDD is again in the first line of the diagram! In its wake, we see the WD800LB (this situation looks strangely familiar to the Disk Transfer Rate).
Apart from the comparison of the drives, you may notice one oddity: the read speed is higher than the linear read speed as measured by WinBench. Well, the numbers are true, we don’t deny…
We like to criticize third-party benchmarks and flayed HDTach, PCMark 2002 and PCMark04 mercilessly, but we also openly talk about our own mistakes.
Well, FC-Test is all right. It accurately measures the time it takes the drive to perform an operation, but the operation system is not always cooperative as to the accuracy of our measurements because all disk operations are performed using the Win API. The inaccuracy arises, if you run FC-Test the way we did.
As usual, it is all about laziness. If we had paid our attention to the higher read speed, we would have corrected ourselves sooner.
So, when running FC-Test, we did the following:
Windows 2000 is not the dumbest OS on Earth, and it probably used its cache (in the system RAM) to produce files requested in Item 3, rather than read them from the drive. Of course, we tried to avoid this by installing only 256MB of memory into our testbed, but couldn’t eliminate the “negative” effect of the OS caching altogether.
There is a solution, however! And we knew about it and even implemented the so-called file-lists in FC-Test. The purpose of a file-list is to restore the test status (to restore the currently opened pattern and locations of the files created on the disk) after a system reboot. By rebooting the system you also flush the Windows file cache as well as the contents of the swap file – the test of file reading is started as if with a blank sheet.
So we are now using the forgotten and newly-revived method of “testing with reboot” and are quite happy with the results as well as with their repeatability. It also turned out that “rebooting” doesn’t practically affect the results of file copying.
Now we return to the results, which are not accurate (caching provided somewhat higher numbers, but it helped each drive equally anyway), but nevertheless interesting. The next pattern consists of average-sized files.
WD740GD is the fastest at reading MP3 files (average size = 4MB or thereabouts), although its advantage over the two Hitachi drives is negligible, comparable to the measurement error. Maxtor 6Y080P0 is also close to the leader.
Maxtor 6Y080P0 is the best of all at reading a set of small files. Seagate ST380013AS did surprisingly well, while the former leader, WD740GD, couldn’t even get into the top three.
Now, let’s see how the drives are copying files within one partition.
Two Maxtors are leaders in copying Install files, but Hitachi HDS722580VLSA80 and Samsung SP0802N are close behind. SP0802N is a real wonder as it manages to compete with 8MB-cached drives with its only 2MB of cache.
The Maxtor drives look preferable at copying large files, but Samsung SP0802N matches their performance. The firmware of this device is especially good for copying large files. I wonder how it will be doing with small files? WD740GB took its deserved place.
Two drives from Maxtor show very good results, but the firmware of SP0802N drive handles this workload well enough, too.
The Maxtors gave up before WD740GD in these tests.
Now we will be copying files from one partition into another.
Two disks from Maxtor did evenly well, Hitachi HDS722580VLSA80 took the third place. The Raptor drive (WD740GD) stumbled in this test – it’s not very stable working with files of different sizes.
When copying files of the ISO type, Raptor regained its leadership, ousting two Maxtors from their position. Samsung SP0802N is grasping hard at the fourth position, in spite of its small cache buffer.
The drives from Maxtor are faster than Hitachi HDS722580VLSA80 at copying average files (4MB).
Maxtor lost its positions in Programs and Windows patterns (that contain files 200KB or smaller). Hitachi HDS722580VLSA80 and Seagate ST380013AS are fast, but Raptor is the best of all. The speed of moving read/write heads around is important for copying small files.
That’s all about NTFS.
The results for the FAT32 file system don’t differ greatly from what we saw in NTFS. So there is no need in closer look at the results, just a few remarks: firstly, the speeds of the drives are slightly higher in FAT32 than in NTFS. Secondly, although the leaders remained the same, the Maxtor drives are somewhat slower in FAT32 than in NTFS and seldom take first places, allowing the Hitachi drives to come ahead. Thirdly, Raptor WD740GD, as well as the Maxtor drives, works better in NTFS.
The comparison of old and new hard disk drives helps us to estimate the progress in the HDD performance. It’s clear that the ex-champion, the DTLA 307075, can only compete with modern devices in applications that require small random access time.
Right now, it is difficult to name a single winner, the best HDD of our days. Western Digital Raptor WD740GD was beyond competition in synthetic patterns for Intel IOMeter, in Fileserver and Webserver, and in WinBench 99, while FC-Test was favorable to two Maxtor drives (6Y080P0 and 6Y080M0) and to Hitachi HDS722580VLSA80. I would also like to single out Samsung SP0802N as it is very fast at copying in the FAT32 file system.
P.S. Your feedback will be welcome, as always. The next roundup of the series is going to be dedicated to 120GB drives – tell us what HDD models you would like to see participating in it?