It all began as usual. The Western Digital company sent us a new hard disk drive model, WD Caviar 800JB. Here it is:
As you may well remember, we have already dealt with a disk from the JB family. It was a 1200JB model, that is, a 120GB HDD. High data density of the 40GB platters of this HDD and the largest cache-buffer among all hard drives with IDE interface (8MB) made it the performance leader. But the storage capacity of JB series (100GB and 120GB) appeared too high and their respective prices turned out really hard to afford. These two factors prevented JB HDDs from becoming really mass products. Well, it might be WD that didn't actually want it and positioned JB series as something "special" meanwhile earning its money on something else. The computer enthusiasts market was soon satiated with JB drives and fast price reduction for smaller solutions led to profits lowering. That's why WD's decision to provide a 8MB cache-buffer for all HDDs with the storage capacity lower than 100GB was quite reasonable. It's also quite reasonable that the 8MB cache-buffer was given to the 80GB HDD - next in the row.
At first, we planned to carry out the performance comparison for three Western Digital HDDs with the large cache-buffer (WD800JB, WD1000JB and WD1200JB). A comparison like that could show us how WD hard disks with different storage capacity and the same cache-buffer perform compared with one another. But all these tests couldn't answer the most urgent question: what the slowest member of the JB family looks against the background of its competitors from other manufacturers (of the same capacity, but with a 2MB cache-buffer). Moreover, as I applied the new testing methodology and used new benchmarks to test the WD800JB, I found myself in the already familiar situation when there's nothing to compare the new disk drive with… :)
In the end, a small article about a better value WD800JB with smaller storage capacity nevertheless being as fast as both WD1000JB and WD1200JB, turned into a comprehensive roundup of hard drives with IDE-interface. It comes on the threshold of the invasion of new HDDs. Let me remind you that we expected new HDDs with 60GB platters, new interface, etc. to arrive in the end of the summer or the beginning of the fall.
But all's for better and the results of this test session will serve us at least twice: for this roundup and to compare today's HDDs with the next generation ones.
The casting for our benchmarks appeared quite impressive. We had well-known Seagate Barracuda ATA IV, Maxtor D740X-6L, IBM Vancouver as well as quite unfamiliar Samsung SP8004H and SV80004H or WD800AB. When choosing the HDDs I tried to follow the "equal capacity" principle so that the performance gap was only determined by the construction and firmware differences.
Testbed and Methods
Our testbed was configured as follows:
- Intel Pentium III (Coppermine) 600MHz CPU;
- ASUS P3B-F mainboard;
- 2 x 128MB PC100 ECC SDRAM by Hyundai;
- IBM DPTA 372050 HDD;
- Matrox Millennium 4MB graphics card;
- Promise Ultra100 TX2 controller;
- Windows 2000 Pro SP2.
Note that now we will use only ATA/100 interface to run the tests. For this purpose we selected a new controller: Promise Ultra100 TX2 with the BIOS officially supporting HDDs with over 137GB storage capacity.
I used the following benchmarks:
- HDTach 2.61
- WinBench 99 1.2
- IOMeter 1999.10.20
- FC-Test v0.3
Before starting the test session all the HDDs were set to "fast" mode by IBM Feature tool utility.
For WinBench tests the arrays were formatted with FAT32 and NTFS as a single logical drive with the default cluster size. The tests were run four times each, the average was taken for further analysis. The HDDs didn't cool down between the tests.
Intel IOMeter test package was performed for the following patterns: WorkStation, FileServer, WebServer, SequentialRead and SequentialWrite.
After a number of successful experiments performed on the external WD drive (see Western Digital External Hard Disk Drive Review), our new HDD test, aka FC-Test, was put on the list. For detailed description of the test goals and methodology, please refer to this article. By the way, you can download it there, too. :)
This test measures average access time, maximum speed of the reads from the buffer, as well as the maximum/minimum/average read and write speeds. From now on I'm going to use Intel IOMeter to measure the maximum read/write speed, as it seems to be better at that. Besides, the average read and write speeds strike me as the most useful values for the user among all the parameters listed above.
The four HDDs from WD with 40GB platters had no rivals in read speed. Surprisingly, WD800BB made of two 40GB-platters showed a slightly worse average read speed and got as fast here as 80GB and 120GB IBM Vancouver drives. The WD1000JB HDD that took part in our tests was to all appearances made of three platters 33GB each and this way fell behind its 40GB-platter mates. But anyway, the average read speed of other 40GB-platter HDDs from Samsung, Seagate and Maxtor was even lower than that of WD1000JB.
Strangely enough, the WD hard drives were not so unanimous when it came to write speed. The two 120GB monsters from IBM and WD notched best. We should note that the effect of the large cache-buffer in the JB HDD family proved far from "magic". It seems that the HDD performance depends only on data density and cache system settings.
WinBench99 is going to yield more than HDTach. First, let's see the average access time of the HDDs.
Maxtor D740X-6L and two IBM Vancouver drives were best here. Not far from them are 7,200rpm drives from WD belonging to both BB and JB families, which showed about the same results. We'd like to single out the WD800AB result. As you see, the average access time of this officially 5,400rpm HDD proved to be the same as that of some 7,200rpm drives.
Well, if Intel can move Celeron CPUs to the Tualatin core, why can't WD equip its AB drives family with motors featuring higher spindle rotation speed? ;)
Second thing we'll do with the help of WinBench test package is to compare the sustained linear read speed of the HDDs:
Well, the picture is similar to what we have just seen in HDTach, when we measured the average read speed. The higher the data density per platter is, the more information the HDD can read from it during one rotation cycle… And again we see the superiority of the HDDs from WD and IBM.
Now let's turn to work with office applications. The diagram displays the results of the HDDs in two integral tests, Business Disk WinMark and High-End Disk WinMark:
The results in High-End Disk WinMark paint to us the following picture:
Two JB HDDs by Western Digital with 80GB and 120GB storage capacity stand nip and tuck on the top. The second place is shared by two IBM Vancouver drives and WD1000JB. Interesting that WD1000JB didn't break away from two IBM HDDs in spite of its four times larger cache-buffer. Could the difference in data density have caused such a great performance gap between the two solutions from the same JB family?
The third place is occupied by the three BB drives from WD. The two Samsung disk drives fell behind in this benchmark.
Business Disk Winmark shows a slightly different picture… The JB HDDs from WD are far ahead, though WD1000JB is the last among them again. The IBM drives are a little better than the BB solutions from WD, although the performance gap between these drives is tiny. The Seagate HDD rose to the occasion, I've never seen Barracuda ATA IV to be so fast. Good results of WD800AB are also worth mentioning. As for Samsung HDDs, we would like to refrain from any comments here.
Let's see how well the HDDs do in NTFS:
The change of the file system didn't provoke any significant changes in the ranks. Well, Maxtor D740X-6L got ahead of Barracuda ATA IV, that's all. As in the previous test, the three JB drives from WD lead the race and are followed by the two IBM drives, and so on…
For those of you who like beautiful pics, I'll show the linear read graphs made in WinBench99:
IBM 120GXP, 80GB (Graph)
IBM 120GXP, 120GB (Graph)
Maxtor D740X-6L (Graph)
Samsung SP8004H (Graph)
Samsung SV8004H (Graph)
Seagate Barracuda ATA IV (Graph)
WD Caviar 800AB (Graph)
WD Caviar 800BB (Graph)
WD Caviar 1000BB (Graph)
WD Caviar 1200BB (Graph)
WD Caviar 800JB (Graph)
WD Caviar 1000JB (Graph)
WD Caviar 1200JB (Graph)
Intel IOMeter: Sequential Read/Write
The main purpose of this pattern is to determine the maximum read speed of the HDD. Actually, HDTach can also provide this information, but there the HDD works with 1MB blocks and we can't be sure that this is the file size with which the HDD can reach the top read speed. That's why the HDD read speed was measured with the data block of changing size (512B-1MB) and the diagram below shows the best results the HDDs revealed.
The results are quite similar to what we saw in Disk Transfer Rate (we do not mean the numbers here, but in the correspondence between the results of the HDDs).
Now, what about writes?
As we can see, streaming data write speed is about the same as the read speed. So why does HDTach show that the write speed is about twice as slow as the read speed? Well, it is its own problem, anyway. ;)
If we compare the results of the HDDs in SequentialRead and SequentialWrite, we'll see that the IBM HDDs got a few steps up on the rating list, but still didn't reach WD800JB.
Samsung SP8004H looked not bad, too. It managed to beat the HDDs from Seagate and Maxtor both in read and write patterns.
Intel IOMeter: WorkStation
Our Workstation pattern was based on the research carried out by the guys from StorageReview. Namely, they studied the statistics of requests addressed to the disk subsystem by the typical office applications in NTFS5. The test results we got with the help of this pattern were unexpected, to say the least of it (hereafter the numbers in the table head denote the request queue depth):
The lowest results of Samsung SV8004H are not a surprise, of course. But the results of Maxtor D740X-6L were simply incredible for me, as well as for you, I guess. As we remember, this drive notched best average access time of all the tested HDDs, but this is not enough to make a good explanation of such high results by Maxtor D740X-6L…
To compare the HDDs with one another somehow, I suggest introducing a total coefficient of HDD performance. The share of HDD speed at a certain request queue depth (Total I/O) will be inversely proportional to the queue depth. So, the Total I/O values will have a greater impact on the final result in case of small request queue (that is, in the mode typical of desktop systems).
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 performance indices, we got according to the formula above, are all provided in the diagram:
Intel IOMeter: FileServer
The pattern name indicates that the HDDs are tested in a mode typical of file-servers. Although IDE HDDs are used in file-servers not as often as the manufacturers might wish :), they do sometimes happen to be there. HDDs from different manufacturers proved to have different performance (what a surprise! ;) ) and the high price of a HDD doesn't necessarily imply high performance…
Maxtor is taking the lead under small workloads. That's natural as this HDD has the lowest average access time value among present-day drives. When the workload is higher, the drives from IBM are coming to the fore and leave others 8% behind. Samsung SP8004H did unexpectedly badly; the request queue depth hardly affects its speed.
As it's hard to predict the average workload for servers, let's take all the workloads as equiprobable. The performance index will be calculated as the average value of Total I/O under all workloads.
Performance = (Total I/O (queue=1) + Total I/O (queue=4) + Total I/O (queue=16) + Total I/O (queue=64) + Total I/O (queue=256))/5
The indices we got according to the formula are presented in the diagram below:
After we had averaged the HDD speeds, the gap between the IBM HDDs and Maxtor D740X-6L turned not that big: only 5%. This trio is followed by the WD host and there's no evident order in their ratings. The 8MB cache-buffer HDDs appeared unexpectedly slower than the BB drive family…
Now we've come to the most interesting part of our review. Here we offer the results the HDDs showed in our new benchmark with a name that speaks for itself. Don't know what you thought about, but FC is the abbreviation for "File Copy". :)
The test is simple. A set of files is created on the HDD. This set is read and copied from one partition to the other and the time spent is measured. For more details on our testing methodology with FC-Test check this article here. Now back to the results.
First, we'll see how the HDDs manage to read and write file sets of different size (meaning the average file size). To make the results obtained easier to analyze, we decided to present them not as the time spent on the operation, but as the average speed in MB/sec.
The Install pattern consists of 414 files with the total size of 575MB. So, the average file size is about 1.4MB.
The results are very interesting… The two IBM HDDs showed highest read and write speed among all the HDDs tested. But the read speed they achieved (~52MB/sec.) is higher than the maximum linear read speed we got in WinBench! Is it possible at all, and if not, why our test produced such a result then? I'm not ready yet to answer my own questions, consider me being deep in thought. :)
To my defense, I would like to say that the test was run three times for each HDD and I got repetitive results (the operating system with its built-in caching may have led us astray…).
We'd also like to stress the high performance of Samsung SP8004H when reading the file set.
The ISO pattern consists of three files with average size of 533MB. So, when creating the files on the disk drive (that is, when write operation is performed) and reading them from the disk drive, the HDD deals with linear write and read.
As we see, the results of this test resemble what we got with the help of IOMeter (read results are especially close to each other).
When working with MP3 files (the average size of these files on my drive was 3.65MB) the IBM HDDs were again ahead:
When working with smaller files (average size ~162KB) the speeds of the HDDs got nearly equal:
But the IBM drives are a little ahead, anyway.
In the Windows pattern the average file size was reduced to 117KB and the writing speed of the HDDs proved about the same:
The drives from IBM proved their superiority once again, though.
Let's make a preliminary conclusion:
- The HDDs from IBM show the best results in reading and writing a set of smaller files while the WD HDDs are best when it comes to larger files.
Let's see how the HDDs copy files. "Copy 1-1" in the diagrams means that the files are copied within one partition while "Copy 1-2" means that the files are copied from one logical drive to the other.
I guess you already know the average file size in every pattern, so let's get to the results.
The JB family from WD show the best performance in copying files within one partition, IBM HDDs are a little behind. But as concerns copying files from one partition to the other, Samsung SP8004H is doing much better than all the others.
The JB family from WD were unrivalled in copying huge files both within one partition and between two partitions.
Both IBM drives and Samsung SP8004H were best working with MP3 files, while the WD HDDs didn't rise to the occasion especially as concerns copying files from one partition to another.
When it came to copying small files the JB series drives from WD took the lead again.
Most interesting… On further decrease of the file sizes the IBM HDDs resumed their leadership again! Real leapfrog, isn't it? :)
The benchmarks results in FAT32 allow me to say that the file-copy test "uncovered" some unknown facts from the private life of HDDs. For example, we got to know that IBM drives are best processing small files while JB drives from WD work excellently with streaming data. Copy tests raised the reputation of Samsung. To tell the truth, I used to regard HDDs from that manufacturer as nothing more than a high-tech toy of the powerful corporation…
I was right about dividing the tests into two sections - FAT32 and NTFS. The file system proved to affect seriously the results of the HDDs in the copy test. The results in one pattern under FAT32 or NTFS differ a real lot. Well, have a look yourself:
Well, the IBM drives won the first two positions, that's not a difference yet… But they are followed not by JB family from WD, but by Samsung SP8004H, Seagate Barracuda ATA IV and Maxtor D740X-6L! And the WD drives are crowded in the end of the list jeering at the clearly unsuccessful Samsung SV8004H. :)
But when it comes to larger files, the WD drives win back the lead in read speed, though the IBM HDDs remain at the top of the list in write speed.
With MP3 files IBM HDDs have no rivals again while Samsung SP8004H and Seagate Barracuda ATA IV managed to get between the drives from IBM and WD.
In this pattern Maxtor overcame the WD drives, too.
On further file size decrease, the brisk Maxtor managed even to become the leader! That's what I call "fast feet"! :)
Interesting that the average-till-now WD1200BB showed rather high results in copying files of this pattern. By the way, note that 120GB drives got grouped in the lower part of the diagram. I guess HDDs with maximum cylinder size get a certain speed bonus when working with average-sized files. Such drives don't need to move read/write heads to read/write such files. They simply switch between the heads.
When working with HUGE files, the JB drives from WD are beyond competition, the BB family looks good as well, though. The IBM drives disappointed me in this mode. Look, they're close to the abyss. :(
But IBM HDDs manage to regain their reputation during MP3 files copying. Samsung SP8004H looks good here, too.
Funny, but in this pattern a better half of the HDDs were faster copying files from one logical drive to the other, not copying them within one partition.
So, what can we deduce from the tests?
Firstly, we have to repeat the phrase said in the introduction:
WD800JB is as fast as both WD1000JB and WD1200JB, but of smaller storage capacity and better value.
It should also be acknowledged that JB drives from Western Digital have no absolute superiority, just like there's no absolute perfection…
Every HDD reviewed here has its advantages and shortcomings, and I hope the growing amount of tests allowed us to look at HDD performance from a new angle.
The analysis of the results shown by the WD drives of different capacity uncovered a certain anomaly. WD 1000BB was faster in some tests than the JB HDDs! As you may guess, I was rather intrigued and carried out a little investigation. A few drives (unfortunately, not all) were still in my possession and I wrote down the firmware version numbers from them:
WD1000BB-00CAA0: 08 Feb, 2002 DCM: HSEACV2CA firmware 16.06V16
WD800JB-00CRA1: 11 Apr, 2002 DCM: HSFHCV2CA firmware 17.07W17
WD1000JB-32CWE0: 31 Jan, 2002 DCM: HSEBBT2CH firmware 22.04A22
WD1200JB-75CRA0: 01 Feb, 2002 DCM: HSEANA2CA firmware 16.06V16
Look at the firmware version of WD1000BB. It's the same as WD1200JB has! It turns out that the same firmware is used in both BB and JB hard disk drive families. That's why these HDDs behave much alike in the benchmarks. Well, from now on we'll have to pay attention to the HDD firmware version as well, it seems…