 |
|
|---|
Revisiting the Caviar WD1000BB and the 'Cuda ATA IV
|
|
Western Digital WD1000BB-SE provided by Western Digital Corporation.
See also Seagate Barracuda ATA IV Review See also Western Digital Caviar WD1000BB Review
Why retest the Barracuda ATA IV? When we initially reviewed Seagate's Barracuda ATA IV we were surprised with the 14.9 millisecond access times that WinBench 99 delivered in light of lower scores reported from a variety of other sites. We combed through these other reviews searching for any mention of Automatic Acoustic Management (AAM) but came up empty. As a result, we assumed the drives were reviewed in their default mode. At the time, Seagate also insisted that it did not plan to offer end users a way to toggle factory-set AAM settings but would rather leave the option to its OEMs. Since then several readers have pointed out that a utility from IBM could be used to disable AAM on the 'Cuda ATA IV (Seagate tells us their own utility is now available through tech support). Doing so brings our measured access times in line with those reported by other sites. Did these sites disable AAM before reporting their results? If so, they didn't report it. Every site in question was based outside North America; perhaps international shipments don't fall under Seagate's new policy of delivering drives with AAM enabled by default. At any rate, it is SR's policy to disable AAM before reporting results. We therefore owe the 'Cuda ATA IV a second look. Further, disabling AAM on the 'Cuda allows us to scrutinize the effects of a reduction in access time on higher-level benchmarks. SR community members hold measured access times in very high regard- most folks consider a one millisecond reduction quite significant. With today's modern firmware and caching strategies, however, just how much difference does one millisecond make? Retesting the Barracuda ATA IV allows SR to gauge the effects of a reduction in access times and its impact on high-level performance. Why retest the Caviar WD1000BB? Long-time readers may remember that SR wasn't too fond of Western Digital's drives when we launched back in 1998. The performance they delivered simply didn't keep up with the competition. In recent times, however, WD has taken the lead when it comes to both capacity and performance. In the wake of Maxtor's recent acquisition of Quantum's hard disk operations, WD stands as the only major manufacturer concentrating its efforts in the ATA sector. As a result, according to company officials, Western Digital has quite a few interesting and unconventional projects on the table. To differentiate their offerings from the competition, one of WD's largest OEMs recently requested an ATA drive with an 8-megabyte buffer. The manufacturer responded by retooling its current flagship, the WD1000BB, with an 8-meg cache. Though the WD1000BB-SE primarily caters to OEM needs, some extra units are available directly through the company's online store for $379... an $80 premium over the standard 2-megabyte BB unit. Aside from its quadrupled buffer, the BB-SE's specs remain similar to the BB. Readers may recall that SR has twice tackled the difference that larger buffer sizes make in the past. The first, involving a pair of Maxtor drives that varied buffer size between 512k and 1024k, proved inconclusive. Our second look contrasted the standard Seagate Cheetah 18LP, equipped with a 1024K buffer, with the 4096K 'A/V' version. Here the Cheetah equipped with the larger cache turned in decidedly better scores. Since then, however, caching strategies, sizes, and SR's own methodologies have evolved considerably. The WD1000BB-SE's 8-meg buffer represents the first increase in ATA buffer sizes since IBM's Deskstar 22GXP and WD's own Expert AC418000 raised the bar to two megabytes nearly three years ago. Testing the Caviar WD1000BB-SE allows SR to gauge the effects of an increase in buffer size and its impact on high-level performance. Let's first take a look at the changes these differences deliver from a low-level perspective.
|
Click here to examine the STR graph for the Caviar WD1000BB-SE
[an error occurred while processing this directive]
|
With AAM disabled, the Seagate Barracuda ATA IV delivers a WinBench 99 measured access time of 13.9 milliseconds, down from the 14.8 milliseconds recorded with default AAM settings. Subtracting 4.2 milliseconds to account for the rotational latency of a 7200 RPM drive yields a measured seek time of 9.7 ms- very close to Seagate's claimed 9.5 milliseconds.
As mentioned earlier, SR readers consider a 0.9 ms difference in access time to be significant. The same difference exists between Maxtor's DiamondMax Plus 60 and the category-leading IBM Deskstar 60GXP. Western Digitial's Caviar WD1000BB-SE turns in a measured access time of 13.5 milliseconds... virtually the same as the standard WD1000BB's 13.4 ms. Subtracting 4.2 milliseconds of latency yields a measured seek time of 9.3 milliseconds, a bit off of WD's 8.9 ms claim. In theory, changes in access times and/or buffer sizes should yield no significant differences in measured transfer rates. And as expected, the 'Cuda IV with AAM disabled (after all, it's the same physical drive) yields no changes. Ditto for the WD1000BB-SE. How do these changes translate into Disk WinMark 99 performance?
|
| ||||||||||||||||||||||||||||||||||||||||||
|
|
The decreased access times gained from disabling AAM on the Barracuda ATA IV bring it a negligible increase in the Business Disk WinMark 99... 0.7% to be exact. Gains in the High-End Disk WinMark 99 are a bit more significant, though at less than 3% still not a difference one would likely notice.
In contrast, the difference in WinMarks turned in by quadrupling buffer size is immense. The WD1000BB-SE exhibits a 25% improvement over the BB in the Business test! This gain builds upon a drive that was already the ATA category leader in this measure. Put into perspective, the BB-SE's score of 10.9 MB/sec trails only Seagate's 15,000 RPM Cheetah X15-36LP... by a scant 2% margin at that! Though improvements aren't quite as breathtaking, they're nonetheless significant in the High-End test. Here the BB-SE builds on the standard WD1000BB (again already the ATA leader here) with a 13% gain, landing itself in the midst of today's top SCSI drives. |
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
Decreased access times almost always yield greater scores in the IOMeter indices, normalized averages of tests that heavily stress random accesses. The lower access times delivered through disabling AAM accelerate the Barracuda ATA IV's IOMeter indices by 6% in File Server and Workstation patterns and by nearly 8% in the Database pattern. These improvements allow the 'Cuda to best Maxtor's DiamondMax Plus 60 across the board though its scores still don't approach that of IBM's Deskstar 60GXP or even the Caviar WD1000BB.
Speaking of WD drives, the WD1000BB-SE's larger buffer doesn't yield it any increases in our Workstation or Database indices... unsurprising considering IOMeter's more random nature. Interestingly, however, the BB-SE realizes a small 4% gain over the BB in the File Server Index. Regular SR readers may have noticed that we've been deemphasizing IOMeter's relative weight in rendering comparative judgments, especially in the case of ATA drives. Why the change? Though our most noticeable activities consist of reviews on hard drives, optical drives, and controllers, much research occurs behind the scenes that never makes publication. Recently these studies have turned to the evaluation of more modern and accurate methodologies to compliment SR's "Testbed3," a project that aims to deploy identical test systems for our three separate review categories. Updated hardware, operating systems, benchmarks, and methodologies will accompany Testbed3. Research indicates that for single-user systems, the ancient WinBench 99 may still hold a lot of merit as an accurate barometer of performance. In addition, these studies have turned in some useful prototypes of measures that will be incorporated in our future testbeds. Let's take a look at one of these tests: IPEAK's RankDisk.
In 1999, Intel released IPEAK SPT v3.0 (Intel Performance Evaluation and Analysis Kit - Storage Performance Toolkit), a comprehensive suite of software designed to assess the performance of various storage subsystems. As just one of IPEAK's many, varied tools, RankDisk plays back standardized sequences of disk accesses drawn from actual use... not unlike WinBench 99's Disk WinMarks. The Business and High-End Disk WinMarks, after all, are merely the disk accesses generated by the Business and High-End Winstone 99 application-level tests. ZD's Winstones play through actual applications scripted to mimic real use. Though in retrospect WB99 exhibits amazing staying power, it nonetheless suffers from age: WB99's patterns were drawn from an old suite of applications running on an old operating system featuring an old file system. IPEAK features a tool called WinTrace32, a program that can capture the access pattern of any given real-world application load. The resulting raw trace file may then be examined (using another component, AnalyzeTrace) or be systematically replayed through RankDisk to comparatively evaluate various driver-controller-disk combinations. The combination of WinTrace32 and RankDisk allows SR to custom-create isolated playback patterns from real-world applications. WinTrace32 requires some intricate OS customization before it can properly trace workloads under the Windows 2000 and Windows XP operating systems. As a result, our initial prototype pattern featured below was drawn from current applications running in Windows NT4 SP6a with the NTFS operating system. This "Typical Use" pattern was captured from an installation occupying about two gigabytes of disk space and represents an actual window of use by yours truly through a variety of light productivity applications. These programs include the various components of Microsoft's Office 2000, Internet Explorer 5.0, a telnet client, an FTP client, ICQ, and a small amount of the first-person shooter Half-Life: Counterstrike. RankDisk presents its results as an average of service times. In this case, "service time" equals the sum of the time required to access the requested data's position and the time it takes to transfer the data. Lower times indicate better performance.
|
| |||||||||||||||||||||
|
|
Though these results are drawn from a prototype trace (final traces will feature a more contemporary OS dealing with data spread over a larger capacity), relevant conclusions may nonetheless be made. Disabling AAM allows the 'Cuda IV to realize a small but significant 5% gain in performance. And in the case of the WD1000BB-SE, its quadrupled buffer propels it 21% ahead of the WD1000BB. The BB-SE's placing is so great that it's trails only the latest 15k RPM drives from Seagate and Fujitsu.
Our File Server, Workstation, and Database IOMeter access patterns play through on all tested drives at queue depths of 1, 4, 16, 64, and 256. To simplify this intimidating benchmark, we introduced the IOMeter indices, a normalized average of results drawn from 16 I/Os, 64 I/Os, and 256 I/Os. Why normalized? Result under a load of 256 I/Os on a given drive are always higher than those drawn under 64 I/Os. Correspondingly, 64 I/O results are always greater than those of 16 I/Os. Normalization simply entails multiplying the lower load scores by a coefficient to bring their weight in line when averaging. Running some raw trace files through IPEAK's AnalyzeTrace suggests that lower-load results are more relevant to single-user disk performance. As a result, we've normalized and averaged results from 1 I/O, 4 I/O, and 16 I/O loads under an "experimental" index. A full sort of all tested drives may be found here. |
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
An interesting shuffle occurs when sorting today's 7200 RPM ATA drives in these new indices. IBM's Deskstar 60GXP, a drive that manages to wrap up top spots in all our deployed indices, finds itself under assault by the WD1000 series. Differences between the BB-SE and BB here remain consistent with the deployed index: identical performance in Workstation and Database patterns yet a measurable increase in File Server performance. The Barracuda ATA IV with AAM off exhibits uniform performance improvements under all three new indices. In a notable twist, however, the 'Cuda IV has trouble keeping up with the competition in these experimental averages even with AAM disabled.
In theory, disabling AAM should result in increased seek noise on the Barracuda ATA IV. Through subjective listening, however, we detect no increase in noise. Without AAM, the 'Cuda remains the quietest 7200 RPM drive that SR's ever tested in both idle and seek situations. The WD1000BB-SE's noise is, not unexpectedly, in line with the BB's. Unlike the 'Cuda, it doesn't plumb the absolute depths of today's noise floors. Even so, the drive's overall profile is hardly objectionable when used in our testbed. Remember, this test system's standard cooling incorporates only the CPU fan on the retail 700 MHz Pentium III and a PC Power & Cooling Silencer 235 ATX power supply. It's hardly a system overrun with the latest in exotic and noisy cooling. Seagate Barracuda ATA IV AAM Conclusion: In the end, disabling AAM results in small performance increases across the breadth of our tests. Despite the improved speed, there's no detectable increase in noise. As a result, performance-minded users may wish to disable AAM to ensure that their 'Cuda delivers its absolute maximum in speed. On the other hand, however, measured increases are indeed small, so those less inclined to fiddle with their drive's settings are hardly missing out on a landmark speed increase. In the end, does a 0.9 millisecond reduction in access time deliver? It's difficult to say. Results suggest that the difference is minimal. And remember, 0.9 ms is 75% of the reduction in access time achieved by moving from a 7200 RPM to a 10,000 RPM spindle speed. At any rate, these minimal margins shed light on why manufacturers such as Seagate and Maxtor ship disks with AAM enabled. Though the 'Cuda IV doesn't exhibit a difference in noise either way, other drives may noticeably benefit from AAM.
An increase in buffer size, properly coupled with a well-conceived caching strategy, yields dramatic improvements in performance. Though Western Digital primarily manufactured the BB-SE for an OEM order, it's clear that they're gauging the market for ultra-high-end ATA disks by offering surplus units to end-users. A survey SR recently conducted comes to mind: |
Western Digital Caviar WD1000BB-SE Conclusion: The BB-SE delivers mind-blowing performance increases over what already was a best-of-class drive. It not only achieves 20+% increases in metrics most relevant to single-user machines but it also yields improvements in file server performance, making it a viable option for those building a server on the cheap. Further, experimental IOMeter indices indicate that the WD1000BB-SE tops the
| |||||||||||||