by Aleksey Meyev
09/04/2008 | 02:34 PM
Some time ago we compared different hard disk drives in terms of their usability in RAID arrays (for details see our first article called 400GB Hard Disk Drives in RAID 0, RAID 5 and RAID 10 Arrays: Performance Analysis). Unfortunately, we have not been able to return to that interesting topic for a while, but now we are ready to renew the series of tests. We will continue from the same storage capacity we stopped at in the last test session, namely 400 gigabytes. In the upcoming reviews we will consider newer HDD models with higher-density platters and higher speeds that have appeared on the market recently.
First we want to tell you about the reasons for such testing and the results of the previous test session.
Apart from pure curiosity, the main reason is to check HDDs out under the specific load of a RAID array. On one hand, the load on each particular HDD within an array is reduced because the HDDs share the same request queue, and sequential requests are likely to go to different disks. On the other hand, the requirements to arrays are higher than to single disks, and the firmware algorithms of look-ahead reading, deferred writing and reordering of requests play a much more important role for them. To make things even more complicated, the RAID controller has some memory of its own and takes part in buffering data and request reordering. And this cooperation between the controller and the HDDs is not always smooth. Long ago a RAID0 array might prove to be hardly any faster than a single disk, not two times as fast, as the user expected.
It is also interesting to see the HDDs work in a RAID5 array which is becoming an ever more popular type of RAID. For the user, RAID5 is a simple way to securing his data and increasing the read speed of the disk subsystem, but for the HDDs this array type is quite difficult because each write request to a RAID5 array translates into two reads and two writes.
We got very interesting results during our previous test session. We could see that NCQ technology really worked and helped improve the performance of a RAID array if supported by the HDDs. This was good indeed because early NCQ-supporting HDDs had behaved rather oddly, often proving to be slower than their NCQ-less predecessors. Another interesting fact was that server-oriented HDDs had no real advantage over their “desktop” counterparts.
Now let’s see what HDDs we’ll be benchmarking today.
Of course, 400 gigabytes is less than half of the maximum storage capacity available today as every manufacturer already offers 1-terabyte disks. However, 400GB drives from older series are still available in shops at a rather modest price. If you don’t need a huge storage space, but want to have a secure or faster disk subsystem, why not use them?
As you can see, six out of the seven disks only differ in specified response time (which is actually not a precise parameter but depends on the HDD’s firmware and the measurement method) and have the same amount of cache memory and the same number of operating surfaces. Well, this makes our tests the more interesting. You’ll see as many as three HDDs from Samsung, two of which differ in the amount of cache memory and the third is a newer model. Western Digital’s drives are exciting, too. Both belong to the server-oriented Caviar RE2 series but the WD4000ABYS was released later and thus has newer firmware.
The following benchmarks were used:
Of course, we built all RAID arrays on the same controller. It was an Areca ARC1220 installed into a PCI Express x8 slot. The controller’s characteristics affect the results, of course, but at least we don’t have to guess which part of the result is due to the HDDs and which, to the controller.
The HDDs were installed into the default cages of the SC5200 system case and fastened with four screws to the bottom of the cage. We decided to benchmark RAID arrays consisting of four disks in RAID0, RAID10 and RAID5 mode.
These modes differ greatly in terms of disk load even when performing the same operations because the second array type differs from the first in using mirroring. As for RAID5, each write request to such an array translates into four requests to the drives in it. Of course, each array was built out of HDDs from the same batch and with the same firmware versions.
The controller was set at the Performance mode for maximum performance during the tests. This mode allows deferred writing and look-ahead reading for both the controller (in its own buffer memory if available) and the disks. Thus we will see if the HDDs’ firmware is optimized for working in a RAID array.
We’ll be referring to the arrays by naming the model of the HDDs they are made of. We’ll call the HDDs their series names. If there are two disks from one series (Samsung’s T133S and Western Digital RE2), we’ll mention the disk’s model name in parentheses.
IOMeter is sending a stream of requests to read and write 512-byte data blocks with a request queue depth of 1 for 10 minutes. The array processes a total of over 60 thousand requests, so we get a sustain response time that does not depend on the amount of cache memory.
There is something odd about the Samsung T166S disks. Judging by the awfully low response time at reading, they are working in the quiet mode (the drive’s actuator is slowed down by changing the shape of the current from meander to sinusoidal). Unfortunately, we couldn’t perform a new series of tests in the ordinary mode after we had found this effect. On the other hand, you will see the effect of this technology on the drive’s real performance at least in comparison with the older-series disks. Take note that the quiet mode has no effect on the response time at writing which is due to the multistep buffering of requests.
As for the other drives, the standings are absolutely the same irrespective of the array type. Hitachi is in the lead, followed by the two drives from Western Digital and Seagate. Samsung’s drives are at the bottom of the diagram, the model with an 8MB cache being somewhat faster than the model with a 16MB cache.
The drives behave in the same manner when writing to RAID0 and RAID10 arrays. The winner Hitachi is followed by the WD4000ABYS (take note of the difference between it and the older WD4000YS). The Seagate has the worst result here. The standings are different for RAID5 due to the complex sequence of actions for performing write requests. As a result, both disks from Western Digital are in the lead whereas the Hitachi proves to be the slowest of all.
In the Database pattern the disk array is processing a stream of requests to read and write 8KB random-address data blocks. The ratio of read to write requests is changing from 0% to 100% throughout the test while the request queue size varies from 1 to 256.
You can view tables with the results by the following links:
We’ll discuss the results for queue depths of 1, 16 and 256.
The HDDs deliver similar performance in a RAID0 array at low loads. The WD4000YS and Hitachi T7K500 are somewhat better than the others when the percentage of writes is high. The Samsung T166S is on the losing side, except at high percentages of writes. It is working in the quiet mode after all. The Seagate is very slow at writing.
The participating HDDs split up under the higher load. Both drives from Western Digital go ahead, the newer WD4000ABYS copying with write requests better. The Hitachi is trying to compete with the leaders but has a slump at 80% writes. There seem to be some flaws in its firmware. Samsung’s three disks are roughly similar to each other, the quiet HD403LJ being as fast as the others and even somewhat faster at reading. The Seagate is good at reading but slows down at writing.
The HDDs behave in the same way under the highest load. Western Digital is in the lead, followed by the Hitachi that still has a slump in the middle of the diagram. The Seagate is good at reading and rather slow at writing.
The HDDs differ more in RAID10 arrays even under low load. WD’s drives are in the lead again, but their graphs begin to fluctuate somewhat at high percentage of writes. As a result, they are eventually outperformed by the Hitachi. Samsung’s HDDs behave differently: the HD400LJ performs write requests slower than the HD401LJ which has a larger cache buffer. The “quiet” HD403LJ is predictably slow at high percentages of reads and accelerates at high percentages of writes but has some fluctuations in performance. The Seagate is good at reading and rather slow at writing, again.
Most of the HDDs speed up at reading when we increase the load. We’ve got the same leaders, though: both drives from Western Digital and the Hitachi. The differences between the firmware versions can be seen easily. The WD4000ABYS is faster than its mate at writing whereas Samsung’s 16MB-buffer drives are equals, leaving their 8MB counterpart far behind. The latter’s performance is somewhat surprising: it looks like its firmware has lost all of its NCQ algorithms.
Western Digital enjoys a larger advantage under the highest load. This brand’s HDDs cope better than the others with the most difficult case when there is the same amount of reads and writes. Again, the Samsung HD400LJ is rather slow at high percentages of reads. Can its cache be entirely dedicated to deferred writing? These are very odd results. We rechecked them but to the same effect. This is obviously a firmware flaw.
Every HDD shows performance fluctuations at certain combinations of reads and writes when working in a RAID5 array. However, Western Digital’s disks are still in the lead while the Samsung T166S takes last place as it has done under low loads in the other arrays. Interestingly, the Hitachi cannot compete with the leaders at high percentages of writes as it did in the other types of arrays.
The graphs smooth out under higher load. WD’s drives enjoy a big lead over the others. Talking about the losers, the Samsung HDDs are worse at reading while the Hitachi is worse at writing. The Hitachi seems to dislike working in a RAID5.
There is nothing new under the highest load. It’s the same as at a queue depth of 16 requests.
The controllers are tested under loads typical of servers and workstations. The names of the patterns are self-explanatory. The Workstation pattern is used with the full capacity of the drive as well as with a 32GB partition. The request queue is limited to 32 requests in the Workstation pattern.
We will be discussing diagrams but you can click the following links for details results and performance ratings:
Well, it’s hard to name a winner here. The Hitachi is in the lead at short queue depths but the WD4000ABYS goes ahead at long queue depths. Take note that its results under high load are far better than those of its predecessor WD4000YS. Samsung’s HDDs are on the losing side, the “quiet” T166S being somewhat faster than its predecessors. The latter two do not differ notwithstanding the different amount of cache memory.
With RAID10 we’ve got the same leaders and losers. But the HD400LJ is now far slower than its opponents, including the same-series model with a 16MB buffer.
There is the same picture with RAID5 as with RAID0. Samsung’s T133S series drives are equals again.
There are write requests now, and Western Digital’s HDDs take top positions at every queue depth. Otherwise, the standings have not changed.
There are no changes when we switch to RAID10. Interestingly, the 8MB version of the Samsung T133S drive is as fast as its 16MB counterpart here.
The Hitachi slows down in the RAID5 array just as we might expect by its Database results. It is now hardly any faster than the slowest HDDs.
The RAID0 arrays show tough competition in the Workstation pattern. The position of the HDD in the table of results depends on the request queue depth. The Hitachi is ahead at short depths whereas the WD4000ABYS prefers longer depths. Take note that the WD4000YS isn’t brilliant at small queue depths, sharing last place with the Seagate. Samsung’s T133S series drives are the slowest at long queue depths.
Western Digital is superior in RAID10, too. The WD4000YS is rivaled by the Hitachi, but the WD4000ABYS is unchallenged. The Seagate takes last place here, being especially slow at small queue depths.
Western Digital’s HDDs are again in the lead. The Hitachi is the worst drive at processing write requests in a RAID5.
When the test zone is limited to a 32GB partition, the competition grows tougher. The Hitachi is unrivalled in a RAID0 at small queue depths. The WD4000ABYS goes ahead at long queue depths. The Seagate is the slowest drive at low loads.
The leader is the same in RAID10 but second place goes to the Hitachi. It is hard to name the loser because the Seagate is closer to the Samsung team at low loads and overtakes them at high loads.
The Hitachi is not so bad in a RAID5 when the test zone is limited to the fastest area of the platters.
Note that the “quiet” T166S has been about as fast as the other drives from Samsung throughout these tests.
IOMeter is sending a stream of read and write requests with a request queue depth of 4. The size of the requested data block is changed each minute, so that we could see the dependence of the array’s sequential read/write speed on the size of the data block. This test is indicative of the highest speed the array can achieve.
You can click the following links for tabled results:
The HDDs split up into three groups when working in RAID0 arrays. Samsung’s T133S series drives and the WD4000ABYS are the fastest here while the WD4000YS is the loser (although the latter achieves its maximum speed sooner than the other drives). The third group show medium speeds.
There are no changes in the standings when we switch from RAID0 to RAID10.
There are certain changes in RAID5. The Samsung HD400LJ achieves its maximum speed later than the other T133S series drive but its maximum speed is higher than the latter’s.
The overall picture changes considerably when we are writing to the RAID0 arrays. We have the same leading trio consisting of the WD4000ABYS and Samsung’s T133S drives, but the HD401LJ has firmware flaws that lead to a performance hit on very large data blocks. The Samsung T166S is now slower than most of the drives.
The HD401LJ has no problems with large data blocks in a RAID10 array but its speed is just slower than that of the model with a smaller buffer.
Every HDD has speed fluctuations on large data blocks when working in a RAID5. This array type is sensitive to the data block size and each HDD model has its own way of treating different data chunks.
The multithreaded tests simulate a situation when there are one to four clients accessing the virtual disk at the same time – the clients’ address zones do not overlap. We will discuss diagrams for a request queue of 1 as the most illustrative ones – the speed doesn’t depend much on the number of applications at a request queue of 2 and longer.
The full results can be seen by the following links:
Interestingly, the only HDD with an 8MB buffer in this review, the Samsung HD400LJ, proves to be the best when reading one thread in a RAID0 array. It gives its leadership up to Western Digital’s HDDs as soon as there are more read threads to be processed, though. The WD4000YS then jumps from last to second place, being only inferior to the newer model. The other HDDs are equals until four threads. Samsung’s HDDs are the worst at multithreaded reading, the T133S series models being somewhat slower than the T166S although the latter has higher response time and lower sequential read speed.
It’s the same as with RAID0. The only thing we can note here is the high performance of the Hitachi when reading two threads.
There are no big changes when we switch to RAID5 except that the Samsung T166S is surprisingly slow at reading three threads.
So, the HDDs’ ability to perform multithreaded reading depends but little on the type of RAID. Is it so with multithreaded writing?
Save for the WD4000YS and Samsung T166S, the HDDs all cope well with writing one thread to a RAID0 array. There are changes in the standings at multithreaded writing, though. As opposed to multithreaded reading, both Samsung T133S drives remain in the lead, followed by the WD4000ABYS. Interestingly, the WD4000YS is the slowest in this test irrespective of the number of threads. Take note that the WD4000ABYS gives way to the Samsung T166S when there are four write threads. Samsung’s HDDs are good at multithreaded writing, at least in RAID0 arrays.
Samsung’s T133S series disks are good in RAID10 arrays, too. They are only inferior to the WD4000ABYS, which is the favorite of this test session, but have fluctuations of performance at certain loads. Particularly, the HD401LJ slows down at two threads (and loses to its 8MB-buffer counterpart) while the HD404LJ accelerates suddenly at four threads. The WD4000YS loses this test, just like it did in the RAID0 array.
Samsung’s T133S series drives are in the lead when writing to a RAID5 array. The HD400LJ is ahead at one and three threads while the HD401LJ, at two and four threads. Interestingly, the latter is faster at two threads than at one. The WD4000ABYS and Samsung T166S are quite fast at multithreaded writing, too.
For this test two 32GB partitions are created on the virtual disk of the RAID array and formatted in NTFS and then in FAT32. After that a file-set is created of the hard disk. It is then read from the disk, copied within the same partition and then copied into another partition. The time taken to perform these operations is measured and the speed of the array is calculated. The Windows and Programs file-sets consist of a large number of small files whereas the other three patterns (ISO, MP3, and Install) include a few large files each.
This test produces too much data, so I will only discuss the results of the Install, ISO and Programs patterns which illustrate the most characteristic use of the arrays. You can click the following links for the full results:
Let’s discuss the NTFS results first.
The two 16MB-cache drives from Samsung cope best with creating large files in a RAID0 array whereas the third Samsung is noticeably slower with every file-set. This is the benefit of a larger cache! Both HDDs from Western Digital join the Samsung team on small files.
The size of the cache buffer is not so important for a RAID10 array and only affects the speed of reading small files. Interestingly, the WD4000YS is slower than the other model from the same series on large files only and equals it on small files.
There is confusion among the leaders when it comes to RAID5. This array type is sensitive to write operations and to firmware algorithms. We can name the losers quite definitely, though. They are the HDDs from Seagate and Samsung.
The Hitachi is in the lead when reading small files from a RAID0 array. Samsung’s T133S series drives are both in the lead on large files. The WD4000YS is the loser irrespective of the file-set.
The same goes for RAID10: we’ve got the same leaders and losers. The WD4000ABYS has got closer to the leaders on large files, though. The Samsung T133S series drives feel somewhat better on very small files, too.
The standings are overall the same when we are reading files from RAID5 arrays.
When copying within the same partition on a RAID0, we cannot name a single winner. Samsung’s T133S drives and the Hitachi are good on large files. The Hitachi becomes the best with medium-size files while the Seagate and the WD4000ABYS are the fastest when reading small files.
There is tough competition in RAID10. The difference between the HDDs is small but we can see that the WD4000YS is the slowest drive with every file-set.
The results are similar again when we are copying files within the same partition on RAID5 arrays. It is only on large files that the Samsung HD401LJ is definitely the fastest while the WD4000YS, the slowest drive. Take note that the speed depends greatly on the size of the file because each write operation takes a few steps to be performed in a RAID5.
When copying from one partition to another on a RAID0, the WD4000ABYS wins on large and small files while the WD4000YS is the slowest irrespective of the file-set.
The same goes for RAID10: one HDD from Western Digital is the winner and another is the loser. Interestingly, the Samsung T133S series drives are the slowest with small files now.
With RAID5 it is on large files that there is a difference in the results. The Samsung T166S is the winner while the WD4000YS is the loser.
Now let’s see what we have in FAT32.
The HDDs from Samsung and Western Digital are competing for top place when creating files on a RAID0.
We have the same leaders when the HDDs are combined into RAID10 arrays. Well, the WD4000ABYS is the only drive from Western Digital to be in the lead here. The WD4000YS is a loser together with the Seagate.
Samsung’s HDDs look the best of all in a RAID5 array. The Seagate and Hitachi produce the worst results. The Hitachi obviously doesn’t like to perform reading in a RAID5.
Samsung’s T133S series drives are again victorious when reading from a RAID0. It is only on small files that they allow the Seagate to take top position.
Samsung’s T133S drives are in the lead in a RAID10 array as well, but they are now rivaled by the WD4000ABYS. The other drive from WD is the slowest through all the file-sets.
We’ve got the same winners and losers when reading from RAID5 arrays.
Copying within the same partition on a RAID0 proves to be a complex test. The drives jump from top to bottom position depending on the file-set. We can only name the worst one here, which is the WD4000YS again.
It’s all normal with RAID10: the WD4000ABYS and the Samsung T133S drives are competing for top place.
Samsung’s HDDs are in the lead in a RAID5 array, too. As opposed to the server tests, they seem to like such loads better.
Samsung’s HDDs are still fast when copying from one partition to another in a RAID0 array but are somewhat slower than the Seagate on large and small files. Both drives from Western Digital sink to bottom places.
We’ve got the typical situation with RAID10: the WD4000ABYS and Samsung’s HDDs are in the lead. The WD4000YS is the loser.
Samsung’s HDDs win this test, too. They enjoy a particular large advantage on small files but are rivaled by the Hitachi on large files.
Here are the HDDs’ data-transfer graphs recorded in WinBench 99:
The next diagrams compare the HDDs in terms of read speed at the beginning and end of the partitions.
There are no surprises here. We’ve seen the same picture in the IOMeter: Sequential Read test. The only odd thing is that the Samsung T166S in a RAID10 array has a surprisingly low speed at the end of the disk. Its data-transfer graph is declining more than the graphs of the other HDDs.
First of all, we’d like to note the superb performance of the WD4000ABYS which was the winner or at least a leader across most of our tests. The older WD4000YS is slower: it feels at ease in most of the server tests but fails in FC-Test. On the whole, Western Digital’s server HDDs have proved that they are indeed better under typical server loads than their competitors almost irrespective of the RAID type.
The Hitachi T7K500 is also good in the server tests, especially under low loads. This is the only disk that challenged Western Digital’s team in this part of the tests. However, this HDD had difficulties working in a RAID5.
Seagate’s 7200.9 series and Samsung’s HDDs are not good in the server tests. The firmware of these drives seems to be oriented at workstation loads. They took last places in the Database pattern where the Seagate showed low efficiency of deferred writing while Samsung’s HDDs showed poor NCQ algorithms, but in some subtests from FC-Test they were as fast as the Western Digital WD4000ABYS.
The Samsung T166S passed our test in the quiet mode. This didn’t affect its performance much, though. This HDD was only much slower than the others under low server loads in the Database pattern. The T166S was about as fast as Samsung’s T133S drives through most of our tests, occasionally outperforming them even.
Now that we’ve done with the past, we are preparing new reviews dedicated to HDDs with larger capacities, higher areal density, and new firmware algorithms.