by Sergey Romanov
06/25/2003 | 09:53 PM
Well, this finally happened! The announcement made last September has finally come true (see this press-release). What am I talking about? I am talking about that notorious Maxtor DiamondMax Plus 9 hard disk drive, which was supposed to have 80GB platters, but turned out to have anything you like except them.
The first 6Y060L0 models to become available in stores were based on 60GB platters. In fact, this is quite logical: what other platter could a 60GB model actually have? So, we asked ourselves this question and started searching for larger HDDs. We found 6Y120L0, but this solution also was built with 60GB platters. We thought: OK, we are dedicated optimists and we will continue our search for 6Y080L0, which should definitely have “proper” platters. No way! This hard disk drive used two 60GB platters, namely three surfaces, where 10GB were simply “blocked”.
This was a blow for our optimism, really. But we didn’t give in. As you know, the DiamondMax Plus 9 HDD family includes 160GB and 200GB models, but doesn’t have any 180GB (60GB x 3) models. Besides, the 200GB model is evidently made differently. Maybe Maxtor decided to save the large-capacity platters for more expensive High-End models? This seemed a reasonable idea, but to check it out we needed a large capacity HDD. So, as soon as we saw 6Y160P0 HDD, we got a couple of pieces for our experiments. After a series of tortures, which we are doing really professionally already, they confessed that there were no 80GB platters in them (see our Roundup: IDE Hard Disk Drives with 8MB Cache-Buffer).
Unbelievable! And what about the announcement stating more than clearly that Maxtor broke all industry records by shipping the world’s highest density HDDs. Have we been cheated on? Is it true that Maxtor was playing an unfair game, while its direct competitors, such as Samsung, Seagate, WD and IBM were manufacturing HDDs on 60GB platters?
Nope. The thing that we have been so desperately looking for for a long time, and which we doubted were a reality at all, has finally started selling in retail. The first HDD revisions were noticed in mid February 2003, about 5 months after the actual announcement. And all those 5 months they were actually selling semi-ready products based on the platter of the same storage capacity as their competitors had. And then Seagate, Maxtor’s No. 1 competitor, released its barracuda 7200.7 with 80GB platters, so in reality Maxtor was not ahead of its rivals. However, Maxtor still was the first one to announce 80GB platters, and while we were trying to figure out what’s going on they started selling the drives based on these platters. Anyway, this time, Maxtor managed to win due to a very smart strategy...
So, what actually happened? What did they do it? Of course, there is no official info answering this question, but I dare share my suppositions in this respect. There is no doubt that Maxtor managed to reach the data density required for 80GB platters. For example, the successfully shipping 6E040L0 model from the DiamondMax Plus 8 family is based exactly on this kind of platter. Nevertheless, the earlier DiamondMax Plus 9 models were based on 60GB and 67GB platters. Maybe, they faced some problems with the mass production of record-breaking 80GB platters or read/write heads for them and trued to find a way out until the problems were eliminated. Now the new HDDs will oust the older models from the market. So, I wonder how greatly the new DiamondMax Plus 9 differs from the older one.
I would like to say right away that the new HDD has an absolutely identical model number as the earlier models, that is why with only a price list in front of you, you will not be able to distinguish between them. However, once you take it in your hands, you will not miss the difference: the new drive is much lighter and “thinner” (there are areas where the packaging is sort of pressed into the drive by about 50% of its thickness, thus making it lighter). It is interesting that Maxtor designed different cases for models with different number of platters. The new HDD was manufactured after mid. December 2002. However, the date doesn’t guarantee that it will be based on new platters.
The new drives have a newer firmware version: YAR41BW0, while the older models had YAR41VW0. At first we thought that it indicates the use of new platters, however, reality turned out much more complicated. We came across 6Y060L0 and 6Y120 models with the new firmware version, but with the old 60GB platters! So, how on earth should we find out how many surfaces a Maxtor drive actually uses? By the drives of the previous generations the number of working surfaces was marked with the last but one digit of the model number, but now there is a “0” there. Maybe keeping in mind the upcoming changes in the platters density, Maxtor decided to give up geometry indication in the HDD model number. But, don’t give way to despair! The more I work with Maxtor drives, the more I discover about them :)
The allegedly lost number finally popped up in the serial number of the drive (the second position). For example, the serial number of the 6Y080L0 HDD looks as follows: Y2TPNN7E, where Y stands for the DiamondMax Plus 9 family and “2” – for two surfaces and two read/write heads respectively. You can find the serial number on the sticker on the upper lid of the HDD case. It is also reported by many utilities working with the hard disk drives.
Because of this confusion with the platter density, Maxtor didn’t make any manuals where they usually listed all exact specifications of the models. The manual hasn’t yet been released, and we can’t wait any longer. That is why our today’s article will have no “Specifications” section. Instead, I am going to simply introduce the today’s testing participants:
The measurements will undoubtedly show that the new 6Y080L0 model has higher Average Access Time than the predecessor. Since the 80GB HDD built of 60GB platters uses not all platter cylinders, the maximum actuator amplitude is 11% smaller, which automatically gives this HDD a significant advantage over the new 6Y080L0. This is exactly the reason why we also included a HDD with smaller storage capacity.
All five hard disk drives feature a 2MB cache-buffer, so they will compete on equal terms with each other. Let the strongest win this battle!
Before we pass over to the actual benchmarks results, let me tell you something very exciting: we have finally found out the notorious thing, which produced wrong results and confused us a lot. This is Maxtor’s brand technology aka Write Verify. As you might remember, this technology first appeared in Maxtor DiamondMax 60/VL30 HDD family and implied that the HDD could read the freshly written data to make sure that they had been written correctly. Of course, this reduced the HDD performance a lot. All Maxtor hard drives sold in those times had the automatic Write Verify set for first 10 boot-ups by default. And then it was automatically disabled. The company has also released a special WVSET utility, which allows enabling and disabling this option.
Since we need the HDD to show maximum performance during the tests, we always disabled this option with the help of Maxtor’s utility, however, it wouldn’t work with the last generation Maxtor drives. Even though the low write speed reveled during the tests indicated that Write Verify was the one to slow it down, there was no way to disable it. Then we started resorting to all sorts of tricks such as turning the testbed on and off ten times before starting the tests, and the like, but they also didn’t always help. The write speed could grow up to the normal level, but the benchmarks results would again be too different. We couldn’t understand anything.
So, one day HDTach benchmark was terminated in the middle and then started anew. The result was like a bolt from the blue: in the interval where we have already run this test the write speed was high, and then after the termination point it dropped again!
Further experiments showed that new Maxtor drives do perform the Write Verify once for each sector on the first write. And now imagine how greatly this fact affected the results of our (and not only our) test sessions!
When testing the read and write speeds, HDTach analyzes not the entire available storage space, but limits itself to 1/65 of it. The remaining surface stays uninvolved, that is why the Write Verify is ready to go any minute. When it comes to WinBench, it uses other parts of the disk surface, but again not all of it. After that Intel IOMeter has to deal with terra-nova, too. Since we do not repeat Intel IOMeter tests several times, because it takes too long, we get the lowered result. So, all benchmarks results suffer from Maxtor’s care in a way.
We don’t quite understand why Maxtor didn’t mention publicly the Write Verify option in the new HDDs, and didn’t modify its utility to support them. Especially, since the HDDs get into reviewers’ hands usually right off the production line, so the Write Verify option shows us its best and worsens the results a lot. Since there are not too many reviewers who can afford to run the entire test session a few times, Maxtor’s results are far not the best and the products usually get not very positive verdicts.
One thing is quite consoling though: now we know exactly where the next underwater rock is, and how we could avoid it. From now on all Maxtor HDDs, which arrive into our lab, are going through “run-in” procedure first. And we are already seeing the positive effect of that. Finally the results of tests run with different controller cards stopped contradicting one another. Unfortunately, we can’t immediately retest all the Maxtor drives, which have already been reviewed on X-bit labs, but the new 6Y080L0 was definitely retested anew after a “warm-up lap”. This is actually why the review appeared a little late.
The testbed configuration looks as follows:
The following software was used:
Before the tests the AAM register of all HDDs was set to OFF position (FAST mode) with the help of Hitachi Feature Tool Utility. For WinBench tests all the drives were formatted in FAT32 and NTFS as one logical drive with the default cluster (to format the drives in FAT32 we used Paragon Partition Manager utility). The tests were run seven times each, the maximum result was taken for the diagrams. The drives didn't cool down between the tests. The tests in Intel IOMeter were run in SequentialRead, SequentialWrite, DataBase, WorkStation, FileServer and WebServer patterns. If you are looking for the detailed description of these patterns, please, see our previous articles.
The competition will start in the notorious HDTach:
Before we pass over to the major parameters comparison, please, have a look at the CPU Utilization rates. The best result here belongs to Seagate Barracuda ATA V, which is also proven by the other benchmarks. And the new Maxtor’s firmware on the contrary, demonstrates high CPU Utilization, close to that by IBM drive. Very interesting...
What do we see in terms of average access time? Exactly, what I have already warned you about: two HDD groups with the same platter configuration. The new 6Y080L0 is slightly behind the predecessor, which indicates that the data organization on the platter remained the same (the same number of servomarks identifying sectors on the track, see our article called Western Digital WD2500JB HDD: More than Drivezilla?! for details).
By subtracting from the measured time the average rotational latency equal to ca. 4.17ms for all tested HDDs, we will get the average seek time.
As you see, the seek time of the HDDs by three competing corporations appeared very similar. IBM drives used to be the leaders here, and now I really don’t know if I should be upset with IBM or happy with Seagate and Maxtor. :) note, that the use of “shorter” platters (three HDDs from the bottom) ensure about 1ms advantage in average seek time.
Well, we have finally come to those parameters, which usually show all the advantages of the increased data density. The read/write rates always grow up as the data density increases, and this case is no exception. IBM however resorted to such technologies as No-id sector format and didn’t use slower internal tracks, that is why only IBM managed to nearly catch up with the new solution from Maxtor in average speeds.
Here we also see the first indication of a new firmware by 6Y080L0 HDD: the read speed from the buffer grew a little slower. But this is not that important, as this parameter doesn’t directly affect the HDD performance.
Now let’s see how the platters of the today’s testing participants are organized. In the diagram below the measured speeds have been transformed into the millions of Bytes per second:
What can we say here? The new platter helped Maxtor to speed up a lot (it is almost 20% faster), but the IBM solution is almost as fast. Of course, the drives with 60GB platters boast better results in the end of the platter. By the way, the read speed grew up to 60MB/sec, which is beyond the UDMA66 protocol potential.
WinBench99 veteran is doing just fine and can still tell us a lot about the “everyday” HDD performance:
WinBench proves that the average access time of 6Y080L0 is a little worse than that of 6Y060L0, while Seagate suddenly appeared much slower. The explanation is very simple: last time we tested a different Seagate Barracuda ATA V. As you see, things are not so stable by Seagate and the first unit we tested was a little faster.
Let’s start our WinMark analysis with FAT32:
The new DiamondMax easily defeats its competitors in professional applications (High-End WinMark), but in “regular” applications it yielded about 5% to IBM. The “old” 6Y0080L0 won the third prize. I dare suppose that it is a different firmware version that we could blame for this lag.
For some reason the new 6Y080L0 won in UDMA100 protocol, while its predecessors, were almost always faster in UDMA133. Let’s find out which particular tests helped Maxtor win.
During Video editing IBM is ahead, then follows Maxtor. To tell the truth, we do not know exactly, what HE: FrontPage actually measures, but here the new solution from Maxtor has almost catch up with the IBM drive. In Microstation there is no progress, and in Photoshop Maxtor improved its performance and left the competitors noticeably behind. In Premiere all HDDs ran equally fast. But during sound processing the new Maxtor showed really brilliant performance. So, I have every right to state that the lazy write algorithms have been significantly improved in the new firmware version. The last test also showed the advantages of the new 6Y080L0, which allows us to state that the work with smaller files has also been polished-off.
But why are we considering professional applications in the FAT32? It’s high time we passed over to NTFS!
In NTFS, the situation has hardly changed. Only the results obtained with ATA/133 controller grew better than that with ATA/100 controller, and Seagate Barracuda ATA V managed to catch up with the slowest Maxtor drive. Let’s go into details now:
Well, the HDDs behave a little bit differently here. For example, in AVS it was really hard for the IBM drive to retain the leadership, and in Microstation test the laurels went to Seagate, which performed impressively fast in the first three benchmarks, but failed to keep this pace in the next four ones. And it is also very interesting that in many cases ATA/133 has no advantage over ATA/100.
Let us try to figure out what has changed for the last few months when Maxtor was “finalizing” its 80GB platter. Firstly, the linear read speed grew up a lot, which allowed catching up with IBM in terms of video processing speed. Secondly, the lazy write algorithms have been significantly improved.
The great job Maxtor did to eliminate the bottlenecks of its algorithms is seen with a naked eye and leads to evident results: the new HDD is almost an absolute leader. Congratulations!
Well, now we will use another powerful tool to test our hard disk drives: Intel IOMeter. This package can bury the reviewer under a ton of data and if used properly it can serve as an irreplaceable info source about the peculiarities of different hard disk drives. But if used improperly... anyway, let’s not talk about sad things now :)
First of all, we will check how well the hard drives cope with reading and writing data blocks of various sizes.
The poorest result in reading belongs to Barracuda ATA V, and the best result in reading small data blocks – to IBM HDD. As the data block size increases, the new 6Y080L0 leaves its predecessors as well as the IBM drive farther behind. In case of 4KB and 8KB data blocks the UDMA133 mode appears a little better than UDMA100 mode.
And here is the first surprise: Barracuda ATA V managed the small data blocks writing best of all. IBM drive managed to catch up with the leader only when we came to 16KB blocks. And Maxtor drive even outpaced it. What a fast fish this Barracuda, when it wants to be it :) As for the influence of different firmware versions, all Maxtor drives performed nearly identically.
The next pattern to check is the longest and the most informative one. It imitates the work with the SQL database. The block size is 8KB, the share of write and read operation varies, as well as the disk subsystem workload.
The results obtained in this pattern are very illustrative. For example, they tell us how well the drives cope with the access to random data blocks.
The fastest here is the drive from IBM, then comes the “old” Maxtor with the same platter configuration, and then comes Seagate’s HDD. The remaining two drives had a much harder time than the three leaders: they had to move their heads at a longer distance. Nevertheless, we could compare them with one another, especially, since this is exactly why we included 6Y060L0 HDD in this test session.
The average access time when reading data blocks of the average size didn’t get any worse, but when writing these blocks it has become much better. And the ratio between the average access time during reads to that during writes shows how well the drive can accumulate and optimize the write requests, i.e. it characterizes the lazy write efficiency in a quantitative way.
This diagram proves my point about SoundForge results mentioned above: the new firmware of the new Maxtor drive has become much smarter and turned out the best among our today’s testing participants. Strange as it might seem, but Seagate’s drive won the second prize, and the very last one in this race appeared IBM. To tell the truth, this situation is not typical at all for IBM products, other models from the same family usually performed lazy writes as efficiently, as the new Maxtor.
Now let’s have a look at the firmware of the drives tested. Since the picture was almost the same for the workload of 1 and 4 simultaneous requests, I decided to show just a few most illustrative diagrams.
IBM drive stands out s lot among these drives due to its very characteristic curl. In case of low writes share it is ahead, but then it yields to Barracuda ATA V and the “old” Maxtor DiamondPlus 9. The new 6Y080L0 outperforms both of them when the writes share grows big.
The higher gets the workload, the more diverse become the performance curves of the tested drives.
Ahead of all is IBM. Seagate is its closest competitor. Then comes the new incarnation of Maxtor DiamondMax Plus 9 6Y080L0 and in the very end go the Maxtor drives with the older firmware version. It is interesting that although the old 6Y080L0 boasted better average access time, it got completely defeated by its successor. We really felt sorry for it during the test. Maybe this drive is too raw and needs some time to run in, as we have already mentioned above.
The ongoing graphs will help us to evaluate the drives scalability. RandomRead pattern comes first:
The reading results rate the HDDs in indirect proportion to their average access time, but as the workload increases, firmware algorithms do not always lead to positive outcome. If Seagate managed to stay close behind the leader, then the Maxtor drives with the old firmware slow down a lot. We were very pleased to see that this issue had been eliminated in the new firmware of Maxtor drives.
This diagram contains averaged data for all reads-to-writes ratios, except two extremes.
Here IBM shows what it is capable of: having started from the very last position this HDD managed to win in the long run. The drive owes its victory to TCQ technology, which implies that it can put aside and combine the requests (including reads) in the way it considers best.
In case the writes share reaches 100% the graphs look absolutely differently. The HDDs with the lazy write function manage the data quite successfully, when the workload is minimal or medium. But starting from 64 simultaneous requests, requests sorting in the OS drivers matters more. Barracuda ATA V proved the best here, while 6Y060L0 appeared the slowest one, as we had actually expected. The interesting thing is that more advanced management of the 6Y080L0 easily made up for the lack of average access time.
Now that we know everything about the HDDs peculiarities, it is even more exciting to check how well they perform in patterns imitating various application fields.
The first here will deal with professional applications. So, imaging a workstation with a hard disk drive formatted in NTFS5, where several programs running at the same time create a pretty intense traffic with a big share of writes in it.
Hm, the difference can only be noticed under a magnifying glass :) One thing is clearly visible though: IBM is far from being the best, Maxtor 6Y080L0 and barracuda ATA V are considerably better. However, the new Maxtor drive managed to significantly improve its performance compared with that of 6Y060L0, that is why if all the drives were working in equal conditions, it could undoubtedly become the winner.
The general rating can be calculated with the following formula:
WorkstationRating = IOps(1) + IOps(2) + IOps(4) + IOps(8)/1,5 + IOps(16)/2
Exactly what we have already said: the difference between the fastest and the slowest drive is less than 6%. But let me assure you, IBM will still feel much slower than Maxtor 6Y080L0 because of considerably lower results under linear workload.
The second application for our HDDs is the work in a File-Server:
The good old Maxtor 6Y080L0 started just excellently, but then lost its fast pace and let the rivals ahead. We were very pleased to see that the new solution had no electronics issues and managed to elbow its way through almost to the very frontline. Though not to the first position....
The server application rating can be calculated as follows:
FileServerRating = IOps(1) + IOps(2) + IOps(4)/1,5 + IOps(8)/2 + IOps(16)/3
IBM is the indisputable No. 1 in Server applications. Seagate also performed very well, but Maxtor drives turned outsiders there. Where is the great server experience inherited from Quantum?
The last IOMeter pattern is also connected with servers, but this time with servers for Internet content. The major difference from the previous pattern is the complete absence of write operations.
Seagate got a bit faster and is running right behind the leader. The new Maxtor 6Y080L0 can’t show all its potential in Intel IOMeter tests, for the already discussed reasons.
The rating in this pattern can be calculated as follows:
WebServerRating = IOps(1) + IOps(2) + IOps(4) + IOps(8)/2 + IOps(16)/3
All in all, the picture is very similar to what we have already seen. The only difference is an even bigger gap between Seagate and IBM on the one end and Maxtor solutions on the other.
What conclusions can we make according to Intel IOMeter tests? The new Maxtor DiamondMax Plus 9 6Y080L0 showed very good results. It is almost completely free from the awkward drawbacks of the previous firmware versions and boasts enhanced lazy write algorithms. Now it is a very well-done HDD, not targeted for multi-user application fields.
When we developed this test set, we were aiming at two goals: to make the copy speed measurements automatic and to measure operations during the copy process separately (such as reading and writing files). We managed to achieve these two goals in the following way. First the test creates files on the HDD according to a certain pattern set by a special script, which provides the number of files to be created and their sizes. Then the saved files are read and copied within one partition and from one partition to another within a tested HDD. Of course, the time required to complete this entire process is measured. The results are given in the tables below:
Since we observed the same tendencies in FAT32 and NTFS, we decided to cut down the number of diagrams and save you some time looking through all of them :)
In terms of write speed, the new Maxtor 6Y080L0 is an absolute winner, while the previous model performed as fast as the IBM drive last time we tested it. In all patterns the new Maxtor drive defeated the predecessors, showing over 20% performance advantage over them in some cases.
However, as it came to reading the new Maxtor managed to do well only when reading large files. The small and medium-sized files remained within IBM’s powers. However, here the new Maxtor drive also defeats all its predecessors.
During copy operations, different protocols start telling on the results. But in all modes except large files copy, the new Maxtor solution is a little faster than the predecessors.
Copying from one partition to another aggravated the situation for the new Maxtor 6Y080L0. in more than 50% of the cases it is defeated completely by the old HDD model.
The new Maxtor DiamondMax Plus 9 proved to be a worthy candidate for your system. Besides the “raw speed” increase due to higher data density per platter, it also showed very well polished-off firmware. And it is very often more important that high physical speed, as you might have already understood from our previous articles. Like in real life: wild physical power can sometimes be not enough, that is when intelligence may help to achieve much bigger heights.
Of course, enhanced firmware algorithms didn’t make this drive a server one, but allowed it to compete with the solution from IBM on equal terms in video and sound processing applications. Our today’s test session lets us state that the new Maxtor DiamondMax Plus 9 6Y080L0 is undoubtedly better than the old model: it is faster in most tests and features one platter instead of two. So, it boasts lower noise level and higher reliability.
One thing is still not quite clear: all the drives from the DiamondMax Plus 9 family acquired the new firmware, but does it mean that all the drives will also start using new 80GB platters? Let’s do a short calculation: 6Y080L0 allows saving one platter and one read/write head. The same thing is true for a 160GB model. 6Y120L0 allows saving one read/write head, and the 6Y060L0 model cannot be redesigned to save anything. Now, do you need any more hints the way the modification of the Maxtor DiamondMax Plus 9 family will take? ;)