by Sergey Romanov
11/15/2007 | 11:36 AM
One year is just a moment for a centuries-old oak tree and a long enough time for a man, but about half a lifecycle for a modern hard disk. Driven by tough competition, the manufacturers have made a point of updating their product lines at least once each six months. The production time of each particular modification is getting shorter, and writing HDD reviews is quite a tricky task. Just when you get enough products to perform a comparative test, there appears a new one that should be reviewed as well. And if you put the review away for a while to perform some other tests, you often return to it to find that it is too late: most of the products are already out of production, replaced with more advanced ones, and are leaving the shops rapidly. That was what happened to our review of 250GB hard disk drives we wanted to publish last summer. There are reasons why this article wasn’t completed then.
So, does it make sense to get back into the past to compare things that will not be? We guess it does. HDDs with a storage capacity of 250 gigabytes currently have the most optimal price/performance ratio. Significant changes have happened only to Maxtor’s products (Seagate’s HDDs are now being sold under this brand) while the Hitachi Deskstar T7K250, Samsung SpinPoint P120S, Seagate Barracuda 7200.9, and some models from Western Digital can still be seen in shops. Among the thirty 250GB products included into this review we have only published the results of the three HDDs from Hitachi before (for details see our article called Way to the Top: Hitachi Deskstar 7K1000 HDD Review).
Our new testbed has a new SATA/PATA controller, so we can offer you the test results of 27 HDDs tested on it.
Having so many products to be tested, we will publish their characteristics in an abbreviated way so that you could just see the difference between them. The results in the tables will be sorted by model number in alphabetic order.
The two generations of Hitachi’s Deskstar HDDs differ mainly in the electronics: the newer generation doesn’t support Tagged Command Queuing and its ATA model has mastered the UltraDMA 133 protocol all the manufacturers, save for Maxtor, used to neglect. The Serial ATA version has an increased data-transfer rate of 3Gbps and supports Native Command Queuing. The increased areal density allowed to reduce the number of platters per drive to two with appropriate improvements in terms of noise and power consumption. The read/write speed has grown up a little, from 60 to 65 million bytes per second at the beginning of the disk.
Maxtor brought forth a multitude of 250GB models. Now that this manufacturer ceased to operate, we want to pay our homage to the respectable brand by including all the tested versions into this review.
The four HDDs with the letter Y in the name are the oldest Maxtors in this review: three 80GB platters and the first-generation Serial ATA interface. They have taken part in many of our previous test sessions. The MaXLine series intended for data storage systems doesn’t differ from the DiamondMax series fundamentally except for an increased MTBF (1 million hours) and a 3-year warranty.
The next generation of Maxtor drives is a highly variegated one: the original issue (6B250S0/R0) was later complemented with the RoHS-compatible versions (6L250S0/R0), and then the models with increased read/write speeds and a second-generation Serial ATA interface appeared (7V/6V250F0). The Serial ATA 2.0 interface provides a twice higher data-transfer speed (up to 3Gbps), hot swap and staggered spin-up features, and allows the drive to process up to 32 commands simultaneously (Native Command Queuing). NCQ is not the distinguishing trait of the third revision, though. The previous revisions of DiamondMax 10 with a Serial ATA interface had it, too.
All of them come with a 16MB buffer (as opposed to an 8MB one in the previous generation) and contain three platters. The 7B250S0 model tested by us is a presale sample with reduced areal density (and, accordingly, with reduced read/write speeds). The 6L250S0 is a retail sample but it has reduced areal density, too (its sequential read speed is even lower than that of the previous generation of Maxtor HDDs). The tested 6B250S0 and 6L250R0 didn’t use the full capacity of the platters. That is, they are full-fledged 300GB HDDs with an electronic capacity limiter whereas the two drives from the server-oriented MaXLine III series do not use one of the six surfaces.
Samsung set a small record with its P120 series: the read speed was higher than 70MB/s at the beginning of the disk. No other manufacturer could offer such a high “linear” speed at that time. Moreover, the Serial ATA version was declared to support a data-transfer speed of 300MB/s and Native Command Queuing. There’s nothing to add more – the rest will be seen in the tests.
Seagate is represented with two product lines in this review. The lifecycle of the Barracuda 7200.8 series proved to be rather short, but it brought about such innovations as a 16MB buffer, NCQ, and an areal density of 125GB per platter. So, the reasons for the hasty release of the Barracuda 7200.9 series with no distinguishing features, except for a data-transfer rate of 3Gbps, are vague. Traditionally since the Barracuda 7200.7 series, the ATA models have a deliberately slowed-down seek, although this is not mentioned in the Barracuda 7200.8 documentation.
There were six models from Western Digital, too. Four of them belong to the first generation of 250GB HDDs (the desktop and server series with ATA and Serial ATA interfaces). The remaining two are newer products: the WD2500KS has updated electronics and a 16MB buffer while the WD2500JS has new electronics and “denser” platters. So, we’ve got an opportunity to check out what is a more important performance-related factor, areal density or the amount of cache memory.
Let’s start out now.
This pattern performs sequential reading and writing in blocks of an increasing size in order to find the maximum speed of each HDD as well as the dependence of speed on the data block size.
As we noted in a previous report, the updated electronics of the Deskstar T7K250 works much better with data blocks smaller than 8KB at sequential reading and writing. The cluster size in NTFS is 4KB by default, so it’s a good thing for Hitachi that its HDDs achieve their maximum speed at such data blocks. The lower read speed of the HDT722525DLA380 in comparison with its ATA mate (and the difference in speed between the two versions of the 7K250) is due to the degradation of the characteristics of the HDDs we noted in our earlier reports. Another sample we tested later on had a max read speed like that of the HDT722525DLAT80, but was anyway slower on data blocks smaller than 4KB, probably due to the Native Command Queuing overhead.
Maxtor’s DiamondMax 10 drives of the first and second version process small data blocks even better. They reach their max speed as soon as 2KB blocks. Strangely enough, the latest DiamondMax 10 generation has a considerably higher maximum speed, but its speed of processing small data blocks is as low as that of the previous generation (DiamondMax 9), and the ATA version is even worse than that.
The two Samsung products have odd sequential read results. They are considerably slower than most of the above-discussed models at reading in small data blocks, and they show a very strange correlation between the read speed and the data block size. Besides, having a very good speed at reading, the ATA model suddenly slows down at writing but improves its results in later tests, surpassing the SP2504C in writing as well. These symptoms suggest an after-write check that we found in Maxtor’s HDDs are few years ago.
The ATA-interfaced Seagate Barracuda 7200.8 acted up as well. It refused to accelerate on data blocks larger than 2KB at sequential reading although it did show a read speed of 66MB/s like the version with a Serial ATA interface. The next generation (Barracuda 7200.9) progressed greatly in terms of its speed with small data blocks but did not match the fantastic results of the Maxtors. By the way, the enlarged 16MB buffer didn’t affect the results of the ST3250624A/AS in this test.
Western Digital doesn’t produce any surprises for us. Four HDDs have identical and quite high speeds of reading while the updated electronics of the WD2500JS/KS is somewhat slower. Since the WD2500JS has increased-density platters, it has a higher top speed which is achieved on 8KB data blocks. All the six models are equally successful at writing.
After we’ve checked out the basic features of the electronics of our HDDs, we can discuss the second type of load – the dependence of speed on the number of simultaneous data streams.
In the diagrams below the results are sorted by the averaged read/write speed for each of the four cases. That is, the models that slow down the most when processing multiple simultaneous streams are at the bottom of the diagram. The numbers show the HDD’s maximum speed of reading/writing a single continuous data stream (in 64MB blocks).
The two drives from Samsung have the highest results (we already noted their record-breaking areal density in the descriptive part of this review). The third-generation DiamondMax 10 is close behind the leaders. Maxtors can also be seen at the bottom of the diagram – the same model but with non-typical 83GB platters. Interestingly, the newer Seagate Barracuda 7200.9 is some 2MB/s slower than the Barracuda 7200.8.
When the HDDs have to read data alternately (for example, from two logical volumes), they form three groups with very different results. The Maxtor DiamondMax 10 didn’t lose anything in speed. All the drives from Western Digital and the Hitachi T7K250 lost half their speed while the “old” Maxtors and the Hitachi T7K250 together with all the HDDs from Seagate slowed down tenfold! The HDDs from the third group are not suited for paralleled load as their look-ahead reading algorithms are intended for one data stream only. You can see that all the manufacturers, except Seagate, took this kind of load into consideration in their newer products, but Maxtor’s engineers did just a fantastic job. Samsung’s HDDs, formally belonging to the same P120 generation, perform in an interesting way: the SP2514N obviously has multi-threaded look-ahead reading (like its predecessors), but the SP2504C with a Serial ATA interface lacks it for some reason. This must be the consequence of the attempt to introduce Native Command Queuing into its firmware.
The picture is different when we test the more realistic situation in which the requests are queued. Most of the HDDs nearly reach their maximum speed by reading one data stream mostly, paying little attention to the second stream. The six HDDs with support for Native Command Queuing behave differently, but four of them couldn’t maintain a high speed for both of the streams. The first-revision Maxtor DiamondMax 10 and its counterpart from the MaXLine III series are the only drives to cope with the task brilliantly: they provide equal performance to both consumers (thanks to NCQ) and don’t slow down (thanks to unique look-ahead reading algorithms). The second and third revisions of the same model show a different behavior, reading the first stream to the detriment of the second stream while maintaining almost the same total speed. Can it be that the NCQ support is disabled in them? We’ll check this out in the next section. Right now let’s examine the dependence of write speed on the number of data streams.
It’s already not surprising to see the Maxtor DiamondMax 10 in the lead but the last places of both generations of Hitachi drives are somewhat unexpected. The best the Deskstar T7K250 could do was to overtake the worst one of the Seagate HDDs. Interestingly, the Barracuda 7200.8 with enabled NCQ is almost as fast as the leaders while the two drives from Samsung are equally good at writing despite their different behavior. It is clear that the firmware algorithms are the main factor for this test. Next goes the amount of cache memory, and only then – the linear read speed.
In the Database pattern the hard disk drive has to process random-address read and write requests. The behavior of each HDD at different request queue depths is analyzed below and then a summary of the test is provided at the end.
The Deskstar T7K250 was the first drive Hitachi implemented Native Command Queuing in, and its operation is visible in the graphs at small percentages of writes. The performance growth is low at a small requests queue depth, however. Write requests are obviously processed bypassing the NCQ mechanism.
At a requests queue depth of 64 commands the Serial ATA version of Deskstar T7K250 has no advantages over the ATA version when there are write requests to be processed. It is only when there are no write requests at all and the queue is longer than 8 commands that we can see a difference between them. It means that NCQ won’t be helpful in most of real-life applications. On the contrary, we can expect a reduction of performance considering the dramatic performance hit the HDD suffers when reading multiple data streams in parallel.
The diagrams for the first generation of Serial ATA drives are almost the same except for the efficiency of deferred writing. There are no surprises here. In our Deskstar 7K400 review we noted that this model was slower in the random write test than the same model with an ATA interface. The latter is almost as fast the HDDs of the newer generation here.
The test of writing multiple streams has shown that these four HDDs have similar deferred writing algorithms, so we can’t expect them to differ greatly in write tests.
The first-generation Maxtor DiamondMax 10 is quite a surprise. We haven’t yet seen such an unprecedented efficiency of deferred writing at a non-zero request queue and such a high growth of performance as the queue depth is increasing. Although being slower than Hitachi’s HDDs at 100% reads and 100% writes, the Maxtor is head above them in all the intermediary modes! Write operations are obviously reordered to achieve highest performance and do not conflict with read operations accumulated in the queue. That’s an example of how Native Command Queuing should be implemented!
This drive from the same series is not that brilliant without NCQ. At a load of one request its graph is identical to the previous model’s, but it shows poor scalability of performance depending on load.
The second-generation DiamondMax 10 lost its unique capabilities for some reason. NCQ shows itself at sky-high loads only.
And the third generation demonstrates minimum differences from the second: NCQ is a little more efficient in case of small queue, but it is still not affecting the performance that much. Let’s see how this Maxtor’s metamorphosis will affect further benchmark results.
Even the previous generation – DiamondMax Plus 9 – turns out faster in most cases even with random requests. They seem to have different lazy write algorithms that do not conflict with reading: in the multi-threaded tests they didn’t yield even a little bit to their younger fellows. I would like to remind you that this brand now belongs to Seagate Technology, all Maxtor solutions have been discontinued and replaced with Barracuda 7200 clones.
Two Samsung HDDs with different interfaces performed very similarly, on the contrary, even despite very diverse attitude to multi-threaded reading in the previous section. Native Command Queuing does exist by the Serial ATA model, but provides very little benefit that is noticeable only during reading of long queues. However, lazy writing doesn’t conflict with any read requests at all here.
Only by comparing two Seagate Barracuda 7200.8 hard disk drives with different interfaces we can finally see Native Command Queuing in action – it proves very smooth and hardly visible. Just like the first revision of Maxtor DiamondMax 10, these HDDs process writes together with reads, however, the lazy writes are very conservative without any data accumulation in the buffer and the requests rearrangement efficiency is minimal.
NCQ became much more expressive in the next Barracuda generation, although, it stopped working with writes. However the models with 16MB buffer lost lazy writing almost completely in case of non-sequential requests, and moreover, the NCQ efficiency was getting almost null. This is very interesting, almost as interesting as Maxtor’s metamorphosis described above. Well, these two companies seem to have found one another, hopefully their jointly developed solutions will surprise us again one day.
Western Digital WD2500JB and its Serial ATA analogue are among the oldest hard disk drives participating in our shoot-out. They have no larger data buffer, no Native Command Queuing support, but nevertheless, they compete pretty successfully with their younger rivals. Showing no remarkable results in two pure modes (only reading or only writing), WD HDDs cope just fine with combined workloads demonstrating confident performance improvement as the percentage of write requests grows.
Modernized electronics of the WD2500JS improved lazy writing almost up to the level of the best competitors’ solutions. WD2500KS equipped with a larger buffer performs almost identically to WD2500JD.
Two server hard drives performed just like their desktop fellows, i.e. we didn’t notice any optimizations for specific types of workload. WD2500SB and WD2500JD graphs coincide and the results of WD2500SD are equal to those of WD2500JB. In other words, we can see three Western Digital firmware versions here with different read ahead algorithms and more enhanced lazy writing of WD2500JS. However, we cannot expect these 6 models to reveal any revolutionary performance differences.
So, let's sum up now.
Most of the existing HDDs do not offer efficient Native Command Queuing. Hitachi didn’t modify the look-ahead reading and deferred writing algorithms well enough, so NCQ lowers the drive’s performance in real applications. Seagate’s HDDs never had multi-threaded optimizations, so NCQ can only be useful in a limited number of tests with multiple competing requests for reading. Samsung’s NCQ implementation provokes many questions, too.
Maxtor was the only developer to do everything right, but then they stepped back for some reason. Perhaps a full-featured support of Native Command Queuing may lead to a loss of data in case of an emergency shutdown of the HDD, but the price of the changes in the newer models from Maxtor is too high. The only good news is that the speed of multi-threaded reading didn’t suffer and Maxtor’s HDDs do not lose anything in performance when NCQ is enabled.
The next three patterns imitate a more complex disk activity in a heavily loaded workstation and two types of servers. We’ll compare the HDDs using a performance rating which is calculated as the total of speeds at different request queue depths with appropriate weights.
The Workstation pattern simulates multiple applications working in parallel. Thanks to a large amount of system memory the OS caches quite a lot of accesses, and the HDD often has to write down only the results.
The first test uses the whole capacity of the hard disk, and the entire Maxtor DiamondMax 10 series is on the losing side, except for the earliest versions with their superb NCQ implementation. The Western Digital drive with updated electronics is the winner. As we noted in the previous section, it has a rather effective deferred writing that does not conflict with read requests. This helps it show high results at small request queue depths. As many as five drives – WD2500KS, Samsung SP2514N, WD2500JD, WD2500JB and Seagate ST3250824AS – are contending for the second place on the podium. They go neck and neck through the entire test, i.e. at every request queue depth. The third place goes to the two HDDs from Maxtor which are accelerating as the load grows.
Full-featured Native Command Queuing helps where classic deferred writing algorithms fail. Strangely enough, none of the Hitachi HDDs has a high speed in this test. The best average seek time and the support for NCQ do not help them get higher than the middle of the list. The two HDDs from Western Digital are just as fast – they had a slump in the graph for the Database pattern.
The identical capacity of the HDDs allows us compare them using a part of their storage space, which largely negates such a parameter as average seek time. So, we run the same test on a 32GB partition.
The reduction of the test zone improved the positions of the Maxtor DiamondMax 10 and Hitachi HDDs, but the overall situation didn’t change. There are but minor changes among the leaders: the WD2500KS together with the Maxtor 6B250S0 and 7B250S0 have got closer to the WD2500JS while the others have fallen behind. The Seagate ST3250624AS (a Barracuda 7200.8 with a 16MB buffer) loses even to its mates with an ATA interface despite having a better average seek time and NCQ. The ST3250824AS, on the contrary, is a step forward in comparison with the Barracuda 7200.8 and looks competitive against products from other manufacturers. So, we can see again how important firmware algorithms are for reaching high performance. A fast seek, high data-transfer rate, large buffer – none of this can compensate imperfect algorithms of look-ahead reading and deferred writing.
The File-Server pattern imitates a heavy load on a data server. Besides a smaller percentage of writes in comparison with the Workstation pattern, it features a larger load range, but this doesn’t prevent the WD2500JS from becoming the winner again. This time it is just a little faster than its pursuers, though.
The competition has become tougher: the Maxtor 6B/7B250S0 have an overwhelming advantage starting with four simultaneous requests. The Seagate ST3250824AS isn’t much worse than the leader at small requests queue depths, and is faster at high loads thanks to NCQ. The Seagate ST3250823AS performs better, too, as its Native Command Queuing is somewhat more efficient. The two drives from Samsung lost their positions, the Serial ATA version being only superior to the dense group of the Maxtors.
The last of the IOMeter tests is a pattern that simulates a Web-server. When calculating the rating, the weights are shifted towards heavier loads. Write operations are missing here.
It is the average seek time that’s important here, and the Hitachi 7K250 has no rivals in this parameter irrespective of the interface. Native Command Queuing doesn’t help the Hitachi HDT722525DLA380 and the Samsung SP2504C, which proved to be the worst drive in this test even. The Seagate ST3250823AS beats the younger ST3250824AS that has been faster in the previous patterns.
Deferred writing is the strong point of Samsung’s HDDs – it is going to be a good factor for them in the other tests.
For this test we format a 32GB partition on the HDD in FAT or NTFS to achieve more consistent results. The diagrams show the best result out of seven runs of the test.
There are a few HDDs in this review that we tested right after we began to use new testbeds. It turned out later than the driver supplied with the Promise Serial ATA II S150 TX2 controller didn’t provide adequate performance in NTFS, so there are two groups of disks in the diagrams, tested with different drivers. The results of the Business Winstone and High-end Winstone tests are combined in one accumulated diagram as they complement each other.
The Maxtor 7B250S0 boasts a very good performance in the total of the two tests. In its group of disks (tested with the older driver) it is ahead of everyone in the High-end Winmark score, notwithstanding its low linear read/write speeds due to less capacious platters. Equipped with the same platters but with updated electronics, the 6L250S0 is only in the middle of its group while the 6B250S0, with the same electronics but “normal” platters, is faster in both Business and High-end Winmark scores. The third modification of DiamondMax 10 with platters of a still higher capacity is just a little slower than the winner WD2500KS, which outperformed not only the newer WD2500JS but also the two Samsung HDDs thanks to its 16MB buffer. As we found out in the IOMeter tests, the algorithms implemented in the SP2504C and SP2514N differ considerably, but their performance is identical in WinBench, and the model with a Serial ATA interface and NCQ is even a little faster than the SP2514N whereas the Seagate Barracuda 7200.8 and Hitachi Deskstar T7K250 are faster in their ATA than in the Serial ATA + NCQ versions. As you can see, this problem is corrected in the next generation of the Barracudas. Interestingly, the two HDDs with a Serial ATA interface from the server-oriented Maxtor MaXLine series show the worst Business Winmark result.
On an NTFS partition the results of the HDDs differ by 50% when tested with different Promise drivers. That’s the explanation of the phenomenal growth of performance of Hitachi HDDs we noted in an earlier review: in fact, the performance of the Deskstar T7K250 is not more than 10% higher than that of the previous series. The first place is now shared by the WD2500KS and Samsung SP2514N. The WD2500JS and the other Samsung contend for the third place. The Maxtor 6V250F0 has lost four positions and found itself behind the Maxtor 6B250S0. As we noted in the IOMeter section, the third-revision DiamondMax 10 works worse with data blocks smaller than 4KB. The ST3250624AS has got six positions higher, obviously due to its 16MB buffer (because the similar ST3250824AS is the last but one in the table of results). The enlarged buffer doesn’t help much to the other HDDs – the Samsung SP2514N, WD2500JS and SP2504C that occupy the positions from second to fourth are equipped with only 8 megabytes of memory.
So, what can we say about the comparison of the thirty HDDs in two file systems in WinBench? We’ve made sure once again that the linear speed is not the main performance-related factor. It becomes crucial only when other factors such as firmware algorithms are the same. If the HDD has got a lucky combination of look-ahead reading and deferred writing, even the amount of cache memory becomes a matter of secondary importance. Otherwise, the buffer size is a decisive factor, but it cannot help outperform a HDD with more advanced firmware.
We formatted the HDDs in NTFS for the PCMark tests. To minimize the effect of the OS on the results, we repeated the test ten times and wrote down the maximum number.
We are becoming more suspicious about the correctness of this test, so we won’t discuss each of its components individually. Let’s see the overall picture instead.
Having the best results in the two subtests with the highest weights, the WD2500KS wins with a considerable lead. Two ATA drives from Hitachi share second place as they have the best results in the Windows boot-up test. The Samsung HDDs haven’t done anything exceptional in PCMark’04, yet take places in the top ten between the Serial ATA models of the second-generation Maxtor DiamondMax 10 series and the Western Digital drives. The Maxtor DiamondMax 10 Plus 9 and all of the Seagate Barracuda did not fight for top places, although the models with a 16MB buffer had an excellent result when copying files. Well, we’ll examine copying and other file-processing operations in more detail in the next section.
FC-Test is indicative of real-life performance of a hard disk drive because it operates with files rather than with pre-recorded low-level access traces. For this test we format a 32GB partition. In this partition files-sets are created, read, and copied. Then we also copy the file-sets into a second partition on the same HDD. We use a total of five file-sets that reflect the folder structure of a real PC.
To avoid overloading this review with similar diagrams and comments we will only discuss the NTFS results. As explained in the WinBench 99 section, the models are divided into two groups depending on the version of the Promise controller driver they were benchmarked with.
The Install pattern imitates a folder with drivers, software distributions and other things necessary to reinstall an OS. It contains a total of 575 megabytes in 414 files.
The Samsung SP2514N copes best with creating this file-set whereas its Serial ATA version fell behind. Maxtor drives of all generations performed well in this test, only allowing the Seagate ST3250624AS with a 16MB buffer and a Serial ATA interface to enter their ranks. Well, for the other HDDs the size of the cache buffer is not crucial. Everything depends on the algorithms implemented in the electronics. The WD2500KS that has won many of the previous tests is not higher than the middle of the list while all the drives from Hitachi and Seagate with an 8MB buffer are at the bottom of the diagram.
The Western Digital WD2500JS/KS, Maxtor 7L/7B250S0 and Seagate ST3250823AS find it an inconvenient operation to read a mixed set of files. The other HDDs are ranked according to their linear speeds.
Copying files is a combination of reading and writing, but you cannot predict the copying performance of a HDD basing on its speed in the read and write tests. This is illustrated by the second place of the Seagate ST3250624AS which hasn’t been higher than eighth until this test. First place goes to the Samsung SP2514N that has won the two previous tests.
Two other models from the Barracuda 7200.9 series, excepting the ST3250824A, have good results, too. The manufacturer must have paid much attention to the copying operation when developing them.
The Samsung SP2504C finds it a problem to transfer data from one partition to another while its ATA relation feels all right at the top of the list. Samsung’s earlier HDD generations used to have high speeds of copying over long distances, but now the SpinPoint P120 with the classic ATA interface is the only model to keep up that tradition. The amount of cache memory separates the old and new Maxtor DiamondMax families sharply, but the 6B250S0 is the only of them to make good use of its 16MB buffer. The WD2500KS doesn’t profit by its enlarged buffer again.
The next pattern is indicative of the maximum speed achievable with each HDD since it contains only three files totaling 1.6GB.
As we already know, Maxtor’s HDDs feel at their ease under sequential loads, and the top of the list is occupied by them. The Samsung SP2514N doesn’t give up, though, winning third place. The old DiamondMax Plus 9 delivers excellent performance, proving again that the areal density and the buffer size are not so crucial for real-life work. The best of the Seagate drives is only 13th. The Hitachi drives proved to be the worst ones in this test.
When reading such large files, almost all of the HDDs, excepting the Seagate ST3250823A, reach their maximum speed. Let’s see what we have when copying within the same partition.
There are not one but two winners here. The Seagate ST3250624AS and the Maxtor 7L250S0 have absolutely identical results, outperforming the potentially faster Maxtor DiamondMax 10 and MaXLine III of the third revision. Having the highest areal density, the Samsung SP2514N comes to the finish only after the WD2500KS that is equipped with 84GB platters. The Seagate Barracuda 7200.8 and all the Hitachi T7K250 drives are on the losing side. The Seagate ST3250824A is against the slowest at copying.
The Seagate ST3250624AS is less agile at copying large files from one partition to another and it lost a few places. The Maxtor DiamondMax Plus 9 looks good as they compete with models that surpass them in terms of areal density as well as memory amount. The two HDDs from Samsung just fail this rather simple test.
Now we reduce the size of the files to the typical size of an MP3 composition. The next pattern includes 271 files totaling 990 megabytes.
It’s something in between the Install and ISO patterns. The two generations of Maxtor HDDs struggle with each other, leaving no chance to their opponents. The Samsung SP2514N is the only HDD that finds itself capable of winning second place from the Maxtors. The other Samsung is among the group of slow HDDs that mostly consists of Seagate and Hitachi products.
Taking the maximum read speed of the drives for the reference point, the reading of MP3 files produced predictable results. The DiamondMax Plus 9 drives are very successful while the Maxtor 7L250S0 fails obviously. The four HDDs from the Seagate Barracuda 7200.9 series are not quite good, either. The WD2500KS and WD2500JS are rather slow, too.
Copying the MP3 file-set proved to be a problem for all Hitachi drives and, to a lesser extent, for the Maxtors. But both Seagate drives with a 16MB buffer follow close behind the leading Samsung SP2514N.
The Seagate ST3250624A triumphs in the Copy Far test, even outperforming the speedy Samsung SP2514N! The Samsung SP2504C, on the contrary, shows the lowest result just like it did with the Install file-set. This seems to be an inconvenient load for the Hitachi drives as well.
And now we will discuss the most interesting case – the processing of small and very small files. The Programs file-set consists of 8504 files totaling 1380MB. The Windows file-set includes 9006 files with a total size of 1060MB.
If you analyze the results of the ISO, MP3 and Install patterns, you can see that the Samsung HDDs prefer to process lots of small files rather than a few large ones, and now the SP2514N shows its best. It is almost 20% faster than its closest pursuer WD2500KS in the Windows pattern! The WD2500JS enjoys a confident third place but there are changes in the lower places. The second and third revision Maxtor DiamondMax 10 drives are losing their positions, allowing to be overtaken by the Hitachi Deskstar T7K250. Seagate’s HDDs are obviously not enthusiastic about this kind of load, delivering the worst results of all.
The Maxtor DiamondMax 10 is better at reading many small files than at writing them. The WD2500JS and WD2500KS are, on the contrary, slow, being even slower than their own predecessors. The Samsung SP2514N wins both of the patterns, and one of the Hitachi HDDs has done well in this test, too. Let’s see what we have in the copying tests.
That’s the moment of truth: the drives that failed the reading and writing tests cannot provide a high copying speed. The only Maxtor drive to look more or less competitive against the teams of Samsung, Seagate and Hitachi is the 6B250S0 model. The others are at the bottom of the list, competing only with the Seagate ST3250824A.
When copying to another partition, the same Maxtor even takes third place, being only inferior to the Samsung SP2514N and Seagate ST3250624AS. We can also see the loser now. It is the Maxtor DiamondMax Plus 9. The Samsung SP2504C is on the losing side, too, yet another time.
We should note that the inadequate results of some HDDs (particularly, the results of the Hitachi HDT722525DLA380 in comparison with the HDT722525DLAT80) are likely to be due to the deterioration of the state of the particular HDD sample during our tests.
This review covering rather old products and also being a kind of a final performance for Maxtor, we won’t make our traditional conclusion about the preferable application of each product. Instead, we’ll point at the strong and weak points of HDDs from different brands, some of which are already corrected in the newer models and some are not.
The Hitachi Deskstar 7K250 proved to be among the slowest HDDs in all of the parameters, save for the average seek time. As a result, they only delivered a higher-than-average performance in IOMeter’s Web-server pattern. The Deskstar T7K250 series is faster in every test, except for IOMeter and PCMark, but its performance is only really better in the multi-threaded reading. The speed achieved is, however, far from the record-breaking performance of the Maxtor DiamondMax 10 and inferior to the results of Samsung’s ATA-interfaced models and to the WD2500KS/SD. The speed of writing is rather low with medium and large files, and Native Command Queuing often lowers the performance to the level of the previous Deskstar generation.
The 6B250S0, i.e. the first-revision DiamondMax 10, looks like the leader among the Maxtor drives we tested. It doesn’t have obvious drawbacks while its implementation of NCQ is just amazing. Unfortunately, the later modifications of DiamondMax 10 are worse in many parameters, especially when processing lots of small files.
The two Samsung SpinPoint P120 drives with different interfaces often had absolutely different results in our tests. The SP2514N would mostly fight for top places and even had no rivals at writing, reading and copying small files, but the SP2504C is rarely higher than the middle of the table of results. Perhaps we had a defective sample, yet we are quite sure that Samsung couldn’t keep the strong points of its products, but spoiled multi-threaded reading if not something else when attempting to introduce Native Command Queuing into its products. The odd behavior in IOMeter’s sequential reading tests was continued in FC-Test where the SP2514N easily won the Windows and Programs patterns. Samsung seems to have developed very efficient caching algorithms.
The comparison of the two Barracuda generations doesn’t reveal Seagate’s work on improving a specific parameter of its products. The ST3250823A is often faster than the ST320824A, but the ST3250823AS is often slower (this may be due to different NCQ implementations). The models with a 16MB buffer seem to belong to a completely different family. Failing in the File-Server and Database patterns, they are very good at copying large files. The obvious drawback of all generations of Barracudas is the significant difference in performance between ATA and Serial ATA models. The former can only leave the last places when copying files from one partition to another.
Finally, Western Digital’s Caviar drives of various modifications surprised us with their stable results, clever optimizations of firmware, and a good speed in such an inconvenient test as reading two disk areas alternately. The WD2500KS won in WinBench 99 and PCMark’04 thanks to its 16MB buffer, and the WВ2500JS was the leader in IOMeter’s Workstation and File-Server patterns, but both had certain problems in FC-Test. The other models had similar results, showing themselves as sturdy mainstream products.
So, each manufacturer has got a lot of work to do yet. In the next part of our comparison of 250GB HDDs we’ll examine modern products from the surviving manufacturers. It’s going to be a very hot competition!