by Nikita Nikolaichev
10/31/2003 | 10:17 AM
Every year the demand for high-speed hard disk drives keeps growing. Although not so long ago the 15,000rpm HDD market was completely dominated by Seagate, now the situation has become absolutely different. There are four HDD manufacturers in this market today, which are struggling really hard to become the king. The first ones to start looking for a market share there were Fujitsu and IBM. As we remember, Fujitsu MAM and IBM 36Z15 hard disk drives were pretty competitive products, so Seagate could no longer retain the monopoly in the 15K HDD market. However, Seagate’s position in this market is still pretty stable, especially since in the early 2003 Cheetah X15 36LP was replaced with a newcomer aka Cheetah 15K.3.
However, Seagate’s competitors didn’t keep their hands in pockets also. In spring 2003 Fujitsu released new MAS 15K drives of the second generation and Maxtor debuted with a new Atlas 15K HDD. IBM, however, quit the race having announced the division responsible for HDD development and distribution a separate company, which later merged with the HDD division of Hitachi Company. The new Hitachi Global Storage Technology Company continued selling old IBM and Hitachi drives under a new company name. They also announced a few new models, including a new 15K HDD aka UltraStar 15K73 on January 6, 2003. Unfortunately, we didn’t manage to get hold of the Hitachi drive, but the other three rivals are here, ready to go :)
The links below will lead you to detailed reviews of all three drives, which you might want to check out before reading on:
And in the meanwhile lets make up a short table comparing the major key features of the drives.
As you see, most parameters of these drives are identical: buffer size, maximum number of platters and heads. The differences lie in the average seek time and average track-to-track.
The curious thing is that Seagate HDD claims to have the highest average access time (3.6ms) and at the same time boasts the lowest track to track time. The shortest average seek time seems to belong to Maxtor drive, although the 3.2ms seek time is specified only for 18GB and 36GB models. The eldest 73GB model is supposed to have 3.4ms already! You probably wonder why I am paying so much attention to these parameters? Well, it is simple.
Once of the principles, which I have always tried to stick to when comparing HDDs performance, was to compare HDDs of the same storage capacity only. Because no matter what various sources will say, the size does matter. Unfortunately, I haven’t yet received the Maxtor Atlas 15K HDD featuring 73GB storage capacity, while the drives from Fujitsu and Seagate participating in our roundup will be exactly of this size. To my disappointment, I will hardly be able to get the desired Maxtor drive really soon, so we will have to put up with that and to compare large HDDs from Seagate and Fujitsu with a small but very fast Maxtor solution, which will not be quite fair to a certain extent.
Our testbed was configured as follows:
To connect the hard disk drives we used Adaptec 29160N controller card with BIOS version 3.10.0 and drivers version 4.10.4002 and Adaptec 39320D controller card with BIOS version 4.10.1 (HOST RAID disabled) and drivers version 1.0. the controllers were installed into PCI64/66MHz slot.
The reviewed drives had the following firmware version:
We used the following benchmarking software:
For WinBench tests the arrays were formatted in FAT32 and NTFS as one partition with the default cluster size. The WinBench tests were run five times each; the average result was taken for further analysis. The HDD didn't cool down between the tests.
To compare the hard disk drives performance in Intel IOMeter we used the FileServer and WebServer patterns:
These patterns are intended to measure the disk subsystem performance under workloads typical of file- and web-servers.
Our colleague, Sergey Romanov aka GreY, developed a WorkStation pattern for Intel IOMeter basing on the StorageReveiw's study of the disk subsystem workload in ordinary Windows applications. The pattern was based on the average IPEAK statistics StorageReview provided for Office, High-End and Bootup work modes in NTFS5 file system and mentioned in Testbed3 description.
The pattern serves to determine the attractiveness of the HDDs for an ordinary Windows user.
Well, and in the end we checked the ability of the drives to work with sequential write and read requests of variable size, and tested the drives performance in DataBase pattern, which imitates the work of the disk subsystem with SQL-like requests.
As usual we will start our discussion with the results of the DataBase pattern. If you are looking for exact numbers, check out the table below containing the results for Total I/O demonstrated by the drives under 5 types of workload (we changed the requests queue depth) in 11 modes with various write requests share (from 0% to 100% with 10% shift).
Now let’s take a look at the testing participants and their performance with U160 controller (Adaptec 29160N) and U320 controller (Adaptec 39320D) in case of different queue depth. We will start with linear workload (one outgoing request):
As we see, the HDDs behave the same way with both controllers. Maxtor drive is the fastest in RandomRead mode, i.e. with no write requests at all. And there is nothing surprising about it actually, because Maxtor drive boasted the lowest average seek time value of all the today’s competitors. However, as the share of write requests grows up, Fujitsu drive catches up with Maxtor and outperforms it in those modes where the writes share is equal or over 80%. Exactly in these modes Maxtor drive looks weaker than its competitors.
Seagate drive appeared as fast as Fujitsu drive in RandomRead mode, however, yielded to the latter in all modes with write requests.
When the workload increased to 16 requests, the HDDs demonstrated their unique individuality. Maxtor Atlas 15K was ahead in case of low writes share, and Fujitsu MAS in case the writes share was really big (>=40%). Seagate HDD outperforms Fujitsu MAS in RandomRead mode and Maxtor Atlas in case of high writes percentage.
Also I would like to draw your attention to the fact that the performance of Fujitsu MAS appeared dependent on the controller type in RandomRead mode. It turned out evidently faster on U160 controller compared with the results on U320 controller.
When we increase the request queue depth, Seagate HDD manages to get ahead of the rivals. It is most likely to be connected with the fact that Cheetah 15K.3 features very deep TCQ buffer (Target Command Queuing), than Maxtor or Fujitsu.
However, the queue depth of 64 outgoing requests is very unlikely to occur on a real server, that is why the advantage of Seagate drive has purely theoretical interest for us.
IOMeter sends a number of read/write requests to the disk array with the queue depth equal to 4. Once per minute the test changes the size of the processed data blocks, so that we can estimate the dependence of the linear read or write speed on the data block size when the test is complete.
For more illustrative discussion I marked with blue and red colors the best and worst results respectively for each data clock size.
No doubt that Maxtor drive is an indisputable leader when working with small data blocks.
Of course, you can see that when the drives work with small data blocks, the Maxtor drive is ahead. Its phenomenal performance has been already discussed in detail in our Maxtor Atlas 15K HDD Review. And in case of larger data blocks the leadership belongs to the Fujitsu drive, which boasts the maximum per track data density of all HDDs tested. The really unique linear read speed graph for Fujitsu MAS has also been covered in great detail in our Fujitsu MAS HDD Review.
Now let’s take a look at a few graphs:
With a U160 controller Maxtor drive is just a little faster when processing smaller data blocks than Fujitsu or Seagate HDDs.
As it comes to U320 controller, Atlas 15K drive appears considerably faster than the competitors during small blocks reading.
Let’s check what happened during writes:
And with writing operations, the situation is just the opposite: Maxtor drive with U160 controller is slower than the competitors when processing smaller data blocks!
Although it performed not bad at all with a U320 controller.
As we have already seen many times: if something appears in one place, then something definitely disappear in the other place. This is the good and bad side of any optimization.
So, it’s high time we tested how fast our drives are in patterns imitating disk subsystem workload:
The highlighted cells in the table show more than clearly that Maxtor Atlas 15K drive is the leader here. moreover, it appeared faster with the U320 controller than with the U160 one. Only Seagate drive managed to grab one of those nice blue numbers for itself, although it succeeded only in case of 256 requests, which is a not very typical workload for a disk subsystem, as I have already said above.
The diagrams demonstrate very well the advantages of Maxtor Atlas 15K against the background of other testing participants. It is evidently faster than Fujitsu MAS and Seagate Cheetah 15K.3 in all modes except the one with 256 outgoing requests, where Atlas 15K yields to Seagate HDD.
I also have to draw your attention to a noticeable performance drop of the Fujitsu MAS drive on Adaptec 39320D controller: the HDD seems to be having some real problems with the U320 protocol.
Now let’s analyze the average performance (Total I/O) of our today’s racers, to see who the winner is:
As you see, Seagate and Fujitsu drives appeared “worth one another” on U160 controller, however, on U320 controller Fujitsu drive was evidently slower. The leading position again belongs to Maxtor Atlas 15K.
Now we pass over to WebServer pattern:
Here the gap between Fujitsu and Maxtor seems to have grown smaller!
Although Fujitsu drive again fell behind the leader on U320 controller…
The ratings diagram also pictured Maxtor’s leadership as well as evident problems of Fujitsu drive with U320 controller. Seagate Cheetah 15K.3, however, managed to get a very high score due to its success in case of 256 outgoing requests.
And now we will compare the HDDs performance in completely different working conditions. WorkStation pattern boasts first, big share of writes, and second, small workload range, which never exceeds 32 outgoing requests.
Therefore, our testing participants behaved absolutely differently:
Of course, the best one here is Fujitsu drive. Under low workload it proved simply brilliant. If we recall the situation in DataBase pattern, then Fujitsu drive worked much better when the writes share was big enough, which actually determined its success in WorkStation.
Under low workloads Fujitsu drive is the leader, and under high workloads (for this pattern, of course) Seagate Cheetah 15K.3 gets ahead. Maxtor drive also looks very nice, though it starts lagging behind the competitors under heavy workloads.
However, since the rating formula for the WorkStation pattern depends much more on the HDD performance under small workloads (the assigned weight coefficient is pretty big), the Maxtor drive managed to outpace Seagate Cheetah 15K.3 in the end!
The formula for rating calculation looks as follows:
Performance = Total I/O (queue=1)/1 + Total I/O (queue=2)/2 + Total I/O (queue=4)/4 + Total I/O (queue=8)/8 + Total I/O (queue=16)/16 + Total I/O (queue=32)/32
Note that the results are included in the rating with a lowering coefficient, which is inversely proportional to the value of this workload. In other words, the final HDD results are determined mostly by their performance under small workloads.
So, it turns out that Fujitsu drive is ahead. The second position belongs to Maxtor, while the Seagate drive, which managed to speed up only under high workloads won only the third prize.
In conclusion have a look at our testing participants in WinBench99. we will start with FAT32:
To make the analysis an easy task for you, I again highlighted the best and worst results with blue and red respectively.
That is why it is no hard task to find out that Maxtor HDD won both integral tests: Business Disk WinMark and High-End Disk WinMark.
How big its victory is you will see on the ongoing diagrams:
Well, we can point out that SCSI HDD from Maxtor is beyond any competition in WinBench tests. Although the IDE drives from this manufacturers have also always been very successful here.
Now we have to see if the situation in NTFS is any different:
And here we get a surprise… The leader here appears… Seagate! While Maxtor HDD turned out the slowest. What an optimization, I should say!
If the situation remained the same for Fujitsu and Seagate, then the results for Maxtor Atlas 15K are incredibly low in NTFS file system.
Ok, let’s now consider the data transfer rates in the beginning and in the end of the drive:
As you see, the “densest” platter belongs to Fujitsu HDD, as the ratio between the data transfer rate in the beginning and in the end of the drive is the lowest (1.22). Seagate’s drive managed to get the highest ratio (1.49).
Here are the linear read graphs for your reference. Click on the graph to open:
Well, our test session for three HDDs with 15,000rpm spindle rotation speed showed that despite the similar claimed specifications, the HDDs performed absolutely differently throughout the entire test session.
Maxtor HDD performed very vividly everywhere. Due to small access time value the drive worked perfectly well in patterns imitating the disk subsystem workload, such as File- and WebServer. At the same time, the synthetic SequentialWrite pattern revealed its low speed during write requests processing. Atlas 15K proved especially slow with U160 controller and small data blocks.
DataBase pattern uncovered one more bottleneck of this drive’s firmware: inefficient lazy write algorithms.
Maxtor’s results in WinBench caught me absolutely unawares: in FAT32 the HDD was rocking, while in NTFS it was dramatically slow.
Fujitsu HDD performed not bad at all: it worked very fast in DataBase pattern, which indicates excellent firmware algorithms. At the same time, the results in DataBase got pretty low when the HDD worked with U320 controller. The same performance drop could be observed in File- and WebServer patterns. Hopefully they will eliminate this drawback in the firmware of mass drives.
Seagate Cheetah 15K.3 yielded to its younger rivals from Maxtor and Fujitsu in most benchmarks. I think that it is the optimization for higher workloads that caused this failure today.
In conclusion I would like to say only one thing: the competition in the high-performance HDD sector has become really intense. And this is very good news for the users. Let the strongest win…