by Nikita Nikolaichev
10/26/2003 | 03:33 PM
As you remember, not so long ago (in April 2001) Quantum’s HDD division was purchased by Maxtor Company. Since Quantum has never had its own production lines (all the HDDs have been produced in Matsushita-Kotobuki facilities), the major negotiable objects of this deal were Quantum’s intellectual property and well-promoted brand names. One of these brands was Atlas. This proud powerful name (according to the ancient Greek myths it was the name of the titan who was holding the sky) belonged to SCSI drives from Quantum. After the merging Maxtor, which used to have only ATA HDDs division suggested that they should also start producing Atlas SCSI drives.
Moreover, Maxtor not just suggested manufacturing older HDD models designed by Quantum engineers, but was also designing new models (using the same engineering team, I assume?). The first SCSI drive to be launched after the merging appeared Atlas 10K IV with 10,000rpm spindle rotation speed, and a little later they also launched Atlas 15K, which we are going to review today.
This HDD with such a high spindle rotation speed is the first experience for Maxtor engineers, that is why we will pay special attention to it and its features.
We had two Atlas 15K HDDs actually. One of them boasted 18GB storage capacity, and the other one 36GB. The latter HDD was an engineering sample sent out for compatibility tests and the like, while the 18GB model was a fully fledged retail unit.
The differences between the two drives can be noticed with a naked eye:
As you may have already guessed, on the left there is a normal HDD and on the right – a sample. Besides the stickers on the upper side of the drive, I would also like to mention the heat-sink plate at the bottom of the retail drive (you can clearly see it on the lower left picture), while the sample doesn’t have anything like that.
It is interesting that the PCBs of both HDDs feature an empty spot for the memory chip (next to the already existing memory chip). I wonder if it was made just in case the company might run out of memory chips of certain size, or if there will be Maxtor Atlas 15K drives with 16MB of cache memory onboard?
However, the HDD specs do not mention anything like that at all. Officially, the HDD features maximum 8MB of cache memory onboard. Anyway, since we mentioned the specs, take a look at the table below to learn more about them:
The most intriguing number is the average track seek time, while the other values are pretty standard for all 15K HDDs of the last generation. The Average Seek value of 3.2ms claimed for 18GB and 36GB models is a sort of a record, I should say, because none of the manufacturers has ever claimed such a small time.
Well, let’s see if it is true.
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 from StorageReview described in the third edition of their HDD testing methodology.
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 drive’s performance in DataBase pattern, which imitates the work of the disk subsystem with SQL-like requests.
We will start with the DataBase pattern. Why with this one? Well, it reveals just perfectly the peculiarities of the HDD firmware, such as tagged command queuing, lazy writing, etc.
As usual. We will analyze the benchmark results under three types of workload:
Under linear workload (queue=1) the mass unit is considerably faster than the sample drive I n case the share of writes is quite big. This allows us to conclude that Maxtor software developers managed to improve the lazy write algorithms in firmware version DT60 quite a lot. Although, the same drive appeared faster in RandomRead mode also.
The second thing worth mentioning is the performance of both drives, which doesn’t depend on the type of the used controller.
Now let’s see what happens as the workload gets higher:
As you see, the performance difference between the mass product and the sample drive appears simply immense. The retail drive is much faster not only in case the share of writes is high, but also… in all other modes! :)
When the queue depth increases to 256 requests, the Adaptec 29160N controller drivers show that they suit better for the mass HDD, as it performed much faster with this controller. The sample drive is evidently slower than the mass product, however when working with the same Adaptec 29160N controller, it managed to show better results than with Adaptec 39320D controller (especially with high writes share).
Well, the results of Atlas 15K in DataBase pattern can be called excellent. Of course, I mean the retail product that we tested here.
Now that we have already tested the HDD stability when processing random data requests, let’s see how well the drives will cope with the sequential requests. We will use IOMeter for this test and will direct a number of requests with the growing address index (we will requests a set of blocks addressed as 0, 1, 2, 3,…, n). The queue depth is set to 4 requests. Once per minute we change the size of the requests blocks, which allows us to find out if the drive favors (or hates) data blocks of a certain size.
Well, let’s first look at reading:
I would like to draw your attention to… Anyway, you will see it much better on the diagram:
In all our previous Ultra320 SCSI HDDs reviews, the smaller data blocks HDD read speed was always considerably lower with an Ultra320 SCSI controller rather than with an Ultra160 one. Ad here the situation is just the opposite! Moreover, not only the retail drive proved faster with an U320 controller, but even the sample one. Then it is not the tester’s mistake, but the peculiarity of Maxtor HDDs, I assume?
Now let’s check how fast the HDD can read info from its cache buffer, and this way we will be able to measure the maximum data transfer rate via this interface depending on the size of the data block transferred:
Maxtor Atlas 15K @ Adaptec 29160N
Maxtor Atlas 15K @ Adaptec 39320D
It is evident that in case of Adaptec 29160N controller the maximum read speed from the drive is limited by the controller bandwidth (PCI32/33MHz). At the same time the burst read speed with the Adaptec 39320D controller, which supports PCI-X bus standard (but is nevertheless limited by the PC64/66MHz bus in our testbed), appeared beyond 200MB/s. This is amazing!
Maybe this phenomenal interface data transfer rate helps the Maxtor drive to process smaller data blocks in SequentialRead pattern? I believe we will have to undertake a few more experiments before answering this question…
And now let’s check out the SequentialWrite:
Strange as it might seem, but Maxtor drives with U320 SCSI interface perform the writes as fast we they do it with U160 SCSI. Another mystery…
The benchmarks in Sequential Read and Write patterns showed that Maxtor Atlas 15K HDD coped perfectly well with sequential requests of the kind. It is especially successful during reads, it performed at its maximum speed with 4KB data blocks already (and as you know, this is a standard size for the NTFS cluster).
Well, now let’s pass over to HDD tests in the server patterns, which are intended for SCSI solutions in the first place.
Let’s check the dependence of the HDD performance on the workload (queue depth).
As you see, until 64 outgoing requests Maxtor drive perform almost equally fast with U160 and U320 SCSI controllers. As soon as the workload increases, the driver optimizations of the Adaptec 29160N start working, which we have already pointed out to you earlier today.
If we make up a performance rating diagram following our standard rules (average Total I/O under all types of workload), we will see the following:
In WebServer the situation is somewhat different:
The sample drive worked evidently faster with U320. The retail drive also appeared faster with the U320 controller under all workloads except the heaviest one with 256 requests queue.
As a result, the HDDs working with U320 controller managed to win under four types of workload out of five and hence got a much better rating.
WorkStation pattern should imitate the disk subsystem workload when it works with regular office applications that is why it differs from the server patterns by the workload range, where the maximum workload makes only 32 requests. By the way, this levels out the chances for U160 and U320 controllers, because the maximum TCQ depth they support is far beyond the maximum outgoing request queue depth.
Of course, the HDD works equally fast with U160 and U320 controllers under small workloads.
When we calculated the performance rating for this pattern, we supposed that small workloads are more typical of a workstation disk subsystem that is why the HDD performance under small workload matters most of all in this formula:
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
The curious thing is that the mass unit working with U320 controller wins again! No doubt, Maxtor drive really works excellent with U320 SCSI interface.
In conclusion we would like to run a few tests from the good old WinBench99. As you remember, Maxtor ATA HDDs are the leaders in this benchmark so far (see our review called Western Digital WD2500JB HDD: More than Drivezilla?!). I wonder if Atlas 15K also boasts some brand name “WinBench99 algorithms”.
Wow, this is great! The performance is insanely fast! We have already see WD Raptor drive run as fast as 16MB/s in Business Disk WinMark (see our review called WD Raptor: First ATA Hard Disk Drive with 10,000rpm Speed). However, Maxtor Atlas 15K demonstrated simply fantastic results in High-End Disk WinMark: 50MB/s!! Although on the other hand, this result is achieved by a HDD with 15,000rpm spindle rotation speed and 8MB cache-buffer, and working with a very fast (and smart) SCSI controller.
In NTFS the performance should definitely be somewhat lower. But how much lower?
Well, it is pretty surprising that Atlas 15K didn’t perform as amazingly fast in NTFS. And we also see that the leader is a sample drive, not the mass one. It looks as if the NTFS optimization algorithms are still under development. :)
In conclusion, have a look at the linear read graphs:
The obtained results showed that the first Maxtor engineers’ experience in terms of 15K HDD design appeared a success. The drive proved very fast in server patterns and especially in DataBase pattern. I also can’t help pointing out its extremely high performance during sequential read and write requests processing, which makes Atlas 15K a great choice for video editing systems.
And of course, I admire the way U320 SCSI interface support is implemented in Maxtor Atlas 15K drive. As you saw, the drive works perfectly well with it.
However, high performance is not enough for making an Enterprise drive commercially successful. Therefore. Maxtor started working on compatibility certification for its SCSI drives by SCSI RAID controller, SCA-chassises and ready data storage systems manufacturers. The current compatibility list is available here.
On my part, I would like to wish Maxtor all the best and good luck in our upcoming 15K HDDs roundup.