Drawing a Bottom Line Under 2005 and a Whole Generation of Storage Drives
This article concludes our comparison of professional Ultra320 SCSI hard drives of the 2004-2005 generation. Now is the perfect time to do it – to look back and overview the recent past on the threshold of the advance of professional next-gen Serial Attached SCSI hard drives.
Our roundup includes representatives of all modern 3.5" lines with the rotational speed of 10,000 rpm and 15,000 rpm. The scope is even larger, as the current market conditions may introduce unexpected new contenders. I mean the desktop WD Raptor hard disk with SATA interface, but with other characteristics of an enterprise level model (10,000 rpm, fast seek time and increased reliability), as well as the tiny Seagate Savvio 10K.1 – the first 2.5" SCSI hard drive with the rotational speed of 10,000 rpm and some other unique parameters, which has been recently reviewed on our web site (see http://www.ixbt.com/storage/savvio10k1.shtml and http://www.digit-life.com/articles2/storage/savvio10k1raid.html).
We already reviewed each representative of the enterprise-level series of SCSI hard drives from the Americans and Japanese – Fujitsu, Hitachi GST, Maxtor, Seagate, and even HP. So you can read the details on each of these series (quite interesting sometimes) in the corresponding reviews (see the links in Tables 1 and 2). In this roundup we shall pool the key data on these hard drives, in order to survey the current state of affairs in this industry. Especially as we can see several different approaches to designing such hard disk drives.
Specifications of modern Enterprise-class hard disk drives
You can have a look at photos of our models at the links below (Table 3). Their basic specifications are published in Tables 1 and 2. The first one contains 15,000 rpm models, the second one – 10K models from Fujitsu, Hitachi, Maxtor, and Seagate.
According to the official specifications, hard disks of the same class (that is with the same rotational speed and interface) have similar parameters. Capacity of the models doubled compared to similar hard drives of the previous generation, being now 36.7, 73.4 and 147 GB for 15K models and twice and large for 10K models. The top 10K model has reached 300 GB (that is it's comparable to top SATA models with the rotational speed of 7200 rpm, though it's several times more expensive than the latter).
However, there is one nuance – how this capacity is reached by the modern generation of SCSI hard drives. The fact is that the majority of manufacturers use a standard formula of heads and platters – Low-End models use 1 platter and 2 heads, Middle-End models – twice as many, High-End models – 4 platters and 8 heads! That is 15K models use 37 GB platters, while 10K models are based on 74-75 GB platters (except for WD Raptor and Seagate Savvio 10K.1, which have 37 GB platters). For the only exception – current SCSI hard drives from Hitachi transgress this rule, the manufacturer uses platters of a noticeably smaller capacity (approximately 30 GB and 60 GB correspondingly), that is Hitachi Ultrastar 15K147 and 10K300 models use 3/2, 5/3 and 10/5 heads/platters correspondingly. Even though all manufacturers offer similar specific (per unit of area) platter densities – about 60 Gbit per square inch, as a rule. Perhaps, that's how Hitachi wanted to shorten the travel of brackets with heads and improve seek times. But unfortunately, the company didn't manage to do it, as the official average seek time of Hitachi hard drives is approximately on the same level as in their competitors (even worse than in Maxtor Atlas, which do not use cutty platters). Moreover, Hitachi drives are noticeably outperformed by their competitors in minimum seek time, though they offer better maximum seek times (cutty platters paid off at least partially).
The fastest platters in maximum sustained (linear) transfer rates (as well as the best seek times) are demonstrated by both series of Maxtor Atlas (10K V and 15K II). That is Maxtor drives lead "in physics" (but they are much heavier than the other models). Although competitors are very close here – 15K models have already crossed the 90 MB/s mark in linear read rate (on outer tracks), they still don't reach symbolic 100 MB/s. Most 3.5" 10K models manage to clear the 80 MB/s barrier, but they still cannot manage 90 (only Maxtor Atlas 10K V is getting close to this figure).
All hard disk drives of this class have 8 MB buffers, except for Hitachi Ultrastar 15K147, which is equipped with the 16 MB buffer. Next-gen SAS drives will all be equipped with such large buffers I guess. ;) What concerns interfaces, along with the good old Ultra320 SCSI, all players offer similar models for 2Gb Fibre Channel (Hitachi even boasts of 4Gb FC models). Besides, most manufacturers have recently started to equip these series with the fashionable and promising Serial Attached SCSI interface (SAS — 3Gb and many other nice features).
Shock resistance of all modern SCSI models is approximately on the same level, manufacturers also increased the reliability index of their enterprise drives to 1.4 million hours. Another good news is that all the latest SCSI drives use fluid dynamic bearing (FBD). This helped reduce the rotational noise (they stopped clanking, which had an immediate effect on the official acoustic parameters of hard drives) as well as improved their reliability.
Among other peculiarities of modern SCSI drives we can mention new faster compact (in terms of PCB room) disk controllers (see the photos), used by all manufacturers. Solutions with high component integration level are preferred in order to reduce the total number of onboard elements. As a result, more compact controller boards (a half or 2/3 of the HDD bottom) are a good form, while the older models used to have controller boards that covered the entire bottom of the hard drive. Another interesting fact - microcircuitry has reached such a high level that it's no longer expedient to design different controllers for 15K and 10K models (it used to be a rule). And now hard disk drives from one manufacturer with different spindle rotational speeds have similar or practically identical controller boards (take Maxtor and Seagate for example). However, we can still come across some rudiments here (for example, Fujitsu and 10K Hitachi models).
A separate mention should be made of packages for 3.5" hard drives from Fujitsu and Seagate (see the photo). These are special plastic boxes: Seagate has been providing them for several years already (for desktop drives), but Fujitsu started doing it not long ago. Fujitsu's boxes are a tad bulkier, thicker (higher) and more "ornate" than boxes from its competitors (Seagate and Samsung). They are initially designed for SCSI drives.
Fujitsu drive in the brand box.
Shock-protection boxes for Fujitsu (on the left) and Seagate drives.
We have picked out the following typical representatives for modern SCSI roundup:
As we can see, hard disk drives in our roundup are mostly represented by middle-capacity models – 147 GB for 10K models and 73 GB for 15K models. The only exceptions are Seagate Cheetah 15K.4 (top model) and two 10K models – Seagate Savvio 10K.1 and WD Raptor – they are represented by top 74-GB models (there are no models of higher capacity in these series yet :)).
Besides, we included two old models – Seagate Cheetah 10K.6 (top model of the previous Seagate series) and "mysterious" HP/Compaq BD14685A26 drive. In actual fact, it's a slightly modified (firmware mostly) Cheetah 10K6, specially manufactured by Seagate for HP, which you can still occasionally meet in retail.
Seagate drives were tested in default Server Mode (adaptive segmentation of cache). Seagate Cheetah 10K.7 and Seagate Savvio 10K.1 drives were also tested in Performance Mode (adaptive segmentation of cache is replaced with fixed 32 segment formatting, see http://www.ixbt.com/storage/seagment.shtml). In this mode the hard disk drives demonstrate noticeably higher performance in simple consumer ("desktop") tasks.
Performance Test Procedure
We used the following testbed configuration to test hard disks:
Note that the motherboard was chosen on purpose. Firstly, Intel E7210 chipset is intended for workstations and single-processor Low-End servers based on Pentium 4 and Xeon (which conforms well with the application area of single SCSI drives of this class), though it will do well for desktop systems (because it's based on the i875P, which we use for testing ATA drives). Secondly, this professional system contains a fully functional (with some reservations though) PCI-X bus (as well as PCI64), provided directly by the chipset (instead of additional bridges). Thirdly, this motherboard houses a modern dual-channel Ultra320 SCSI controller — Adaptec AIC-7902B, connected to PCI64 bus. Fourthly, this SCSI controller fits our purposes much better (testing single hard drives) than a more expensive one with onboard memory, because Adaptec seems the most popular (at least in our country) brand of SCSI controllers. No additional memory in this controller allows to measure HDD performance in the "purest" form, unaffected by possible caching of the controller and its data stream processing algorithms.
The hard drives are mounted firmly on the metal racks of a chassis and cooled by a small front panel fan for a 5-inch bay (some of the drives would get overheated in a couple of hours without additional cooling). SCSI drives are plugged to the system (one at a time!) via a short 20cm cable, in order to minimize phase incursion at high transfer rates of the Ultra320 interface (this is perhaps the principle scourge of modern SCSI, which keeps this parallel bus from developing into a higher frequency area), that is to obtain the best transfer rates possible along the SCSI bus in our test system. The controller (non-RAID mode) uses Adaptec driver 184.108.40.206 dated September 2004. The tests were carried out under MS Windows XP Professional SP2. The hard disks were tested not partitioned (in Intel Iometer, H2Benchw and AIDA32 tests) as well as partitioned and formatted by regular operating system tools depending on a test type: one NTFS partition of maximum size for testing the average access time and plotting the read speed graph in WinBench 99 and two NTFS or FAT32 partitions of the same size for other tests (WinBench Disk WinMark 99, ATTO Disk Benchmark test, multithreaded read/write, and the Adobe Photoshop CS test). Each NTFS partition occupied half of the entire disk capacity (that is the second partition started exactly from the second half of the disk). FAT32 partitions were 32768MB, the first one starting at the outer edge of the media (on the "fastest" tracks), and the second — exactly from the middle of the media volume. NTFS and FAT32 cluster sizes were selected by default — 4 and 16 KB correspondingly. Before the tests, we warmed the hard disks for 20 minutes using a utility with active random access.
Results of the physical parameter tests
Linear transfer rate graphs (reading) are on the images at the following links:
And the diagram contains maximum, minimum, and average transfer rates measured in WinBench 99.
We should note the following: Maxtor and Fujitsu hard drives demonstrate the best transfer rates in each class. These hard disks achieve such results due to the so-called adaptive formatting, which allows to squeeze maximum performance from each platter retaining a proper reliability level by adjusting formatting density for each head in a hard disk. Adaptive formatting can be seen clearly, if the linear read graphs are plotted with the resolution higher by 100-1000 times than in WinBench 99 (and other tests) and if they are not smoothed (see the picture – periodic oscillations of Maxtor and Fujitsu reading rates indicate platters with different data density). Data density on different platters in one and the same drive may differ by 10-13%!
High resolution graph fragments for 15K-rpm drives.
Diagram 1 also shows that 10K Seagate models (unlike Cheetah 15K.4) are noticeably outperformed in this parameter. What concerns Savvio 10K.1, it's significantly outperformed by the other 10K models and its transfer rates are approximately on the level of last-year 7K models with 80 GB platters. It's not surprising, considering the smaller diameter of its platters.
The contenders also differ in Ultra320 SCSI interface transfer rates, but this time Seagate is at advantage, its platters demonstrating a tad higher results in different programs.
However, such small differences in interface transfer rates are insignificant in case of an effective buffer.
15K models are naturally faster in the average access time, Maxtor Atlas 15K II noticeably outperforms its competitors, which are approximately on a par here. The Maxtor drive also leads among 10K models, having outperformed even Savvio, which seek rates theoretically must be of the same class as in 15K drives (platters of the same diameter as in 15K models). Two-platter WD Raptor also fairs well here, though it does not stand out against SCSI drives. Seagate's average access time, measured in Performance Mode under Windows, grows considerably. Readings in a special Intel IOmeter pattern verify these results (see the diagram below). But it turns out that not all SCSI drives demonstrate adequate reading results in the IOmeter pattern for random access in minimal blocks.
Additional information is provided by the comparison of the average access time measured separately for reading and writing under Windows — we may try to evaluate the efficiency of lazy write procedures and caching of written data in the buffer by how much the average access time drops in writing compared to reading. (Of course, the average write access time in this case does not reflect the real physical property of hard drives! It's just a parameter measured by some Windows tests to evaluate the efficiency of caching write data in a buffer.) Here are the results of two different tests — H2Bench and IOmeter.
The results of both tests are nearly identical — Maxtor Atlas drives are again in the lead (in caching data for writing), though Seagate drives are getting close to Maxtor, while Savvio suddenly (does it?) demonstrates outstanding results – on the level of 15K models! Unexpectedly low results are demonstrated by Hitachi and WD Raptor drives here.
We shall omit the bulky and very interesting ATTO Disk Benchmark results here to save room, you can get these results in corresponding product reviews. And now we proceed to HDD performance tests in applications.
Performance in applications
At first, let's find out how well the hard disks are optimized for multithreaded operations. We traditionally use NBench 2.4 tests for this purpose, where 100MB files are written on the drive and read from it in multi-threaded mode. Here is a diagram with average results in this test. You can find the detailed test results for each pattern on a separate web page.
Maxtor's leadership (only Hitachi 10K300 is faster) in writing in several concurrent threads is doubtless. Cheetah 15K.4 and WD Raptor also look good here, while the 10K drive from Fujitsu and the 15K drive from Hitachi failed this test, being outperformed even by Savvio. But in multi-threaded reading the situation has changed considerably – Maxtor's place is taken by Fujitsu drives (Hitachi 10K300 is still the leader); both new Seagate Cheetah models fair well. However, the 15K model from Hitachi also fairs well this time, having significantly outperformed the main group of contenders, including WD Raptor and both Maxtors.
The old but still popular Disk WinMark 99 tests from WinBench 99 (they evaluate performance of these enterprise hard disk drives in desktop tasks) also reveal obvious leaders – Fujitsu, Maxtor and WD. The other SCSI drives are not optimized for such tasks – due to their server-oriented firmware.
Recent complex benchmarks for desktop performance evaluation in PCMakr04 and C'T H2BenchW use "replaying" previously recorded HDD activity tracks in corresponding applications and measure completion times of each track, the total average result is calculated afterwards.
According to the total score in PCMark04 disk tests, the fastest SCSI drives for desktop and other consumer tasks are Fujitsu, though Maxtor and WD are only slightly behind. Seagate Cheetah 15K.4 is also in good books here, while the other Seagate drives may compete for the place "under a desktop lamp" only in Performance Mode (but they are still outperformed by good 7K ATA models).
C’T H2bench overall performance test makes the desktop picture even more prominent — Fujitsu and Maxtor are much faster than the other SCSI drives here. WD Raptor managed to outperform all 10K SCSI models without exception. The leaders are more than three times as fast as the outsiders!
Differences in Adobe Photoshop swapping performance are much smaller among our contenders (below 1.5 times). Seagate Cheetah 15K.4 takes up the lead, having outperformed Fujitsu and Maxtor drives of the same class. Among 10K models, top three places (without any preferences) are taken by Fujitsu, Maxtor and Seagate Cheetah 10K.7 drives, if the latter is used in Performance Mode. As in many other desktop tests, the outsider is Seagate Savvio 10K.1 (but it can get close to the middle group in PM).
Intel Iometer tests
Enough of wearing our SCSI drives out with the tasks, which are not quite typical of their profile. Let's proceed to more typical profiles. We also use special patterns in Intel IOmeter to imitate hard disks operating in various "serious" applications. At first — traditional popular patterns, offered by Intel and Storagereview.com — imitation of various servers (DataBase, File Server, Web Server) and a Workstation. Detailed performance/queue depth results can be seen in the diagrams on a separate web page (as well as in the commented graphs, for example, in reviews at http://www.ixbt.com/storage/fujitsu15k.shtml, http://www.ixbt.com/storage/fujitsu10k-mat.shtml and http://www.ixbt.com/storage/savvio10k1.shtml). The diagram below contains geomean results of three server patterns (without weight numbers), as HDD positions in each of the three patterns are rather close and reflect the general firmware optimization trends for the current physical characteristics.
It turns out that 15K models have a definite leader – magnificent Maxtor Atlas 15K II, which also fairs well in consumer tasks. But the desktop leader (from Fujitsu) is replaced by Seagate Cheetah 15K.4 and Hitachi Ultrastar 15K147, having left the former leader far behind. The leader outperforms the majority of 10K models by 1.5 times, which is only natural considering the difference in spindle rotation speeds and platter diameters.
The leader among 10K models is again Maxtor Atlas. But this time Fujitsu is at the back of the back. Hitachi Ultrastar 10K300 also fairs well. But Seagate Cheetah 10K.7 is evidently out of favour, being outperformed even by its predecessor - Cheetah 10K.6 (and its reincarnation from HP). Nevertheless, Seagate Savvio 10K.1 is good news here - it unexpectedly outperformed many "sterling" 10K models, even despite a significantly smaller linear transfer rates and approximate parity in average access times – it may be the effect of its well-optimized firmware. Just note that WD Raptor is an outsider here: though TCQ (not quite popular in SATA controllers) can raise performance by 5-10% in these tasks, it will not save the Raptor from being an outsider to modern SCSI drives in this case. The same concerns Seagate drives and PM (it just deteriorates the results).
The situation in the workstation pattern is only slightly different. Category leaders are the same here (Maxtor Atlas, Cheetah 15K.4, and Fujitsu MAT), but modern 10K Seagate models yielded much ground and are outperformed even by the Raptor. Only PM allows them to outperform the latter.
Now it's turn for our own IOmeter patterns, which are also quite indicative of enterprise drives, because even in "deeply professional" systems a lion's share of hard drives' operations fall on reading/writing large and small files and sometimes file copying. As the access pattern (at random addresses across the entire hard disk) is more typical of server systems, our contenders demonstrate high results for these "simple desktop" (on the face of it) patterns of reading and writing files.
Detailed results are published in the diagrams on a separate page (as well as on the graphs in the above mentioned reviews). And in this summary diagram we shall publish geomean performance results of these hard disk drives.
The leaders are well known from the previous chapter of our roundup – these are models from Fujitsu and Maxtor (however, the situation in each pattern is less homogenous, the Japanese is victorious mostly due to its excellent results in reading and writing large files, where Maxtor is significantly outperformed). Old WD Raptor and Seagate Cheetah 10K.6 models also fair well here, outperforming many new 10K models due to their caching optimizations. Seagate Cheetah 10K.7 and Savvio 10K.1 are outsiders, though PM may come in handy and add up to 30% of performance.
Speaking of details, Fujitsu drives demonstrate excellent results in reading large files (e.g. archives, large databases, audio, video, large photos, etc) — they are ultimate winners in reading and writing large files, having outperformed both Maxtor and Seagate. Their performance almost doesn't depend on the queue depth. It suggests an idea that these Fujitsu hard drives will be good in stream data servers (video-on-demand, video monitoring, and many others) and similar tasks, requiring efficient reading and writing of large files and data streams. But Fujitsu drives are less impressive in reading and writing small files (remember weak access times for reading and writing), while they compete with Seagate models in reading (being outperformed by Maxtor in non-unit queue depths) and even outperform the former leader (Maxtor) in writing. Quite decent performance results both for servers as well as desktops. Maxtor drive is the best at copying large and small files (it demonstrates a great advantage over its competitors), but Fujitsu is on guard and will not let pass any model except for Maxtor.
In the defragmentation test Maxtor Atlas drives are again ultimate winners (see the details on a separate web page), but Fujitsu models are "in great form" here as well, noticeably outperforming the other hard disk drives.
And finally, Maxtor is evidently the fastest drive in the streaming read/write pattern with large and small blocks (for example, such loads are quite typical of a video server). Fujitsu is close on its heels. Breakaway from outsiders is huge, only WD Raptor can fill it, having outperformed SCSI drives from Seagate and Hitachi (see the details on a separate web page).
Acoustic noise and power consumption
As I have already noticed above, objectively lower rotational noise of the latest-gen SCSI drives is promoted by fluid dynamic bearing (FDB). The difference from the old Seagate Cheetah 10K.6 (or its HP-clone) is huge (at least with my "naked" ear). But we should note that new Seagate drives (Cheetah 10K.7 and especially Savvio 10K.1) are quieter than their modern competitors. Subjectively, all 15K models were identically noisy in active seek and without it (special measurements are required for precise evaluation). By the way, active seek in current SCSI drives has generally become a tad quieter than in their predecessors.
Our contenders demonstrate similar power consumption and heat dissipation, these characteristics evidently depend on a model capacity. Additional cooling is mandatory for almost all of them, lest they should overheat (but in certain conditions junior SCSI drives from Seagate can operate actively for days remaining just warm without additional cooling). However, we shall publish a separate article, devoted to a detailed analysis of power consumption and heat dissipation in these and some other professional hard disk drives.
That's one of the most important parameters for SCSI drive users (sometimes to the detriment of performance and price). But unfortunately, we cannot evaluate it objectively on our own for an obvious reason (by the way, all hard drives tested in this roundup have been working perfectly, at least for several days of incessant testing). And it would be wrong to use third-party statistics in this case. That's why we just establish a fact that all modern SCSI drives somewhat improved their declared reliability properties (for example, MTBF has grown to 1.4 million hours). Five years of warranty actually cover the entire expedient life span of such hard disk drives.
It's time to draw conclusions on our roundup of modern SCSI drives. Out of doubt, their main merit is doubled capacity — up to 147 GB in 15K models and up to 300 GB in 10K ones. Building high-performance professional data storage systems with such capacities is not just facilitated, but becomes economically sounder. Transfer rates of such drives have noticeably grown as well, of course (the linear read rate has come close to the cherished 100 MB/s threshold). These drives are rather efficient and relatively quiet compared to their predecessors. Their declared reliability also got better. SCSI is still alive and kicking, hopes for its near death are grossly exaggerated, though professional serial interfaces are becoming increasingly popular so that we cannot already imagine product lines without them.
What concerns the most thoroughly researched issue – performance of these drives in various tasks, Maxtor products are ultimate winners in both categories, demonstrating the best results under serious professional load and competing with the leaders (from Fujitsu) in consumer tasks. That was why some time ago we awarded Maxtor Atlas 15K II and Maxtor Atlas 10K V with our Original Design prize. By the way, the same award was given to the innovative Seagate Savvio 10K.1.
It would be a pity if the excellent SCSI drives from Maxtor (descending from the unforgettable Quantum) would be sacrificed on the sly to SCSI and SAS drives from Seagate as a result of the recent merger of Seagate and Maxtor. On my soul! The industry of professional data storage devices will lose much, if all those fruitful designs from Maxtor are just forgotten under pressure of Seagate's marketing managers.
Modern SCSI hard drives from Fujitsu are quite interesting – they are ultimate performance leaders in many tasks, which are more typical of desktops rather than servers and storage systems, but they demonstrate lower results under purely server loads (they are sometimes outperformed by Hitachi and Seagate models). That's why you should have a crystal clear idea how you want to use these drives, if you choose them. Multimedia content servers and workstations seem a decent application for them. Especially as Fujitsu has been recently dumping prices for SCSI models. Hitachi and Seagate drives make sense only if you plan a purely active server operation, as in most cases their performance slumps noticeably under simpler loads. In such cases WD Raptors often offer a better price/performance ratio. If you plan on using 10K Seagate drives for such tasks, you'd better set them to Performance Mode (just a mouse click in Seatools Enterprise).
By the way, you are advised to have a closer look at Seagate Savvio 10K.1, if your hard disks should take up as little space as possible (1U and blade servers as well as 2U storages) – these are promising hard disk drives, a couple of them even in a simple RAID can outperform a larger 15K model (which also consumes more power, is heavier and noisier). See, for example, http://www.digit-life.com/articles2/storage/savvio10k1raid.html)!
In conclusion I want to emphasize that long gone are the times when performance superiority of a given professional hard disk drive was governed almost solely by its physical properties (transfer rates, access times, interface transfer rates, buffer size…) Lately we have often found that products with similar specifications may demonstrate cardinally different performance in various applications. That is physical properties become less important for HDD leadership than the expertise of programmers, who code and optimize firmware versions. We'd like to thank those who do a great job of it and do it with regard to various users. :)
We express our gratitude to Maxtor and Seagate for the provided hard disk drives
Alex Karabuto (firstname.lastname@example.org)
February 9, 2006
Write a comment below. No registration needed!