July 12th, 2006 by eugene
SATA in the Enterprise - A 500 GB Drive Roundup
The enterprise storage market continues to evolve towards a new paradigm. In the past, high-performance applications were served by arrays of 15,000 RPM drives while 10,000 RPM units continued to serve as workhorse units for less intensive real-time applications.
As always, cost is a key factor that drives change. In addition to the price of drives themselves, expenditures relating to housing, powering, and cooling arrays must also be considered. As a result, the industry is moving to a more segregated approach where high-performance 15,000 RPM units service data where speed of retrieval is of utmost importance while slower yet larger and more cost-effective 7200 RPM drives store the bulk of data where a somewhat more leisurely retrieval does not significantly affect productivity.
Thanks to interoperability afforded by the newer SAS standard, SATA drives have stepped into the spotlight. With the ability to easily integrate into serial SCSI infrastructure, enterprise-class SATA drives have enjoyed increased attention from the big three American drive manufacturers.
Seagate, Maxtor, and Western Digital have all entered the fray with SATA units specifically tuned for the enterprise sector. While leveraged from consumer-class SATA designs, these differentiated models undergo tests under different workloads, often enjoy longer factory burn-in cycles, are rated for longer mean times between failures, and are backed by a more business-oriented 5-year warranty.
Let us take a closer look at three 500-gigabyte units that squarely aim to seize the burgeoning nearline enterprise sector where cost and capacity rather than sheer IOps drive the market.
As always, tests on these three contenders were conducted using StorageReview's Testbed4 suite of measurements. NCQ remained enabled for all tests, though figures comparing performance with NCQ on and off may be found towards the end of the review. Lets turn to the figures!
Access Time and Transfer Rate
For diagnostic purposes only, StorageReview measures the following low-level parameters:
Average Read Access Time- An average of 25,000 random read accesses of a single sector each conducted through IPEAK SPT's AnalyzeDisk suite. The high sample size permits a much more accurate reading than most typical benchmarks deliver and provides an excellent figure with which one may contrast the claimed access time (claimed seek time + the drive spindle speed's average rotational latency) provided by manufacturers.
Average Write Access Time- An average of 25,000 random write accesses of a single sector each conducted through IPEAK SPT's AnalyzeDisk suite. The high sample size permits a much more accurate reading than most typical benchmarks deliver. Due to differences in read and write head technology, seeks involving writes generally take more time than read accesses.
WB99 Disk/Read Transfer Rate - Begin- The sequential transfer rate attained by the outermost zones in the hard disk. The figure typically represents the highest sustained transfer rate a drive delivers.
WB99 Disk/Read Transfer Rate - End- The sequential transfer rate attained by the innermost zones in the hard disk. The figure typically represents the lowest sustained transfer rate a drive delivers.
For more information, please click here.
In an interesting finish, all three 500-gigabyte contenders weigh in with identical 13.8 millisecond random read access times. Accounting for the 4.2 ms rotational latency of a 7200 RPM spindle leaves the units with a measured seek time of 9.6 ms. WD specs an 8.9 ms seek for the RE2 and thus misses its mark by 0.7 ms. Maxtor and Seagate maintain progressively more ambitious claims of 8.5 ms and 8.0 ms respectively and thus miss by a wider margin. It should be noted, however, that all three newer units lag previous-generation drives in random read scores. The MaXLine Pro manages to pull ahead a bit when it comes to write scores. The NL35.2, not unlike the older MaXLine III, appears to undergo additional read verification after every write and as a result yields a rather sluggish score.
The 500 GB RE2 leads the pack when it comes to sequential transfer rates. Its outer-zone figure of 73.8 MB/sec improves upon the score turned in by its predecessor by a healthy 14%, as one would expect when moving from 100 GB to 125 GB platters. Maxtor's lineup also shows progress with the newer drive edging into 70 MB/sec territory. The Seagate NL35.2, however, trails the group at just 63.6 MB/sec. While a drop in transfer rates between the NL35.2 and its predecessor is not surprising given the newer drive's lower (125 GB vs 133 GB) platter densities, it remains nonetheless disappointing that Seagate's offering fails to keep up with the 125 GB/platter competition.
It is important to remember that seek time and transfer rate measurements are mostly diagnostic in nature and not really measurements of "performance" per se. Assessing these two specs is quite similar to running a processor "benchmark" that confirms "yes, this processor really runs at 2.4 GHz and really does feature a 400 MHz FSB." Many additional factors combine to yield aggregate high-level hard disk performance above and beyond these two easily measured yet largely irrelevant metrics. In the end, drives, like all other PC components, should be evaluated via application-level performance. Over the next few pages, this is exactly what we will do. Read on!
StorageReview uses the following tests to assess non-server use:
StorageReview.com Office DriveMark 2006- A capture of VeriTest's Business Winstone 2004 suite. Applications include Microsoft's Office XP (Word, Excel, Access, Outlook, and Project), Internet Explorer 6.0, Symantec Antivirus 2002 and Winzip 9.0 executed in a lightly-multitasked manner.
StorageReview.com High-End DriveMark 2006- A capture of VeriTest's Multimedia Content Creation Winstone 2004 suite. Applications include Adobe Photoshop v7.01, Adobe Premiere v6.5, Macromedia Director MX v9.0, Macromedia Dreamweaver MX v6.1, Microsoft Windows Media Encoder 9.0, Newtek Lightwave 3D 7.5b, and Steinberg Wavelab 4.0f run in a lightly-multitasked manner.
For more information, please click here.
The RE2 series regresses slightly when executing tasks found in the SR Office DriveMark 2006, trailing the older 400 GB drive by a 3% margin with a score of 763 I/Os per second. It is enough, however, to best the MaXLine Pro by 5%. Maxtor's newer drive performs identically to the older unit in the Office DriveMark. Bringing up the rear is the NL35 family. While the second-generation unit's Office DriveMark of 508 IOps improves upon the original drive's score by 6%, it nonetheless lags the MaXLine Pro by a whopping 43% margin.
All three 500-gigabyte units finish quite similarly in the SR High-End DriveMark 2006. Here it is Maxtor's drive that leads the trio at 543 I/Os per second. In a marked turnaround from its dismal performance under the Office DriveMark, the NL35.2 squeezes by WD's drive to place second in the 500 GB pack. All three units, however, lag behind WD's older 400 GB RE2 by margins of at least 8%.
Three decidedly different entertainment titles cover gaming performance in StorageReview's test suite.
FarCry, a first-person shooter, remains infamous for its lengthy map loads when switching levels.
The Sims 2, though often referred to as a "people simulator," is in its heart a strategy game and spends considerable time accessing the disk when loading houses and lots.
Finally, World of Warcraft represents the testbed's role-playing entry; it issues disk accesses when switching continents/dungeons as well as when loading new textures into RAM on the fly.
For more information, please click here.
Ironically, it is when running games, applications these enterprise-class drives are hardly marketed towards, that the newer 500-gigabyte units start to muscle their way ahead of their predecessors.
The 500 GB RE2 tops WD's older 400-gig unit by a 6% margin when replaying SR's Farcry drive access script with a score of 704 IOps. The MaXLine Pro and NL35.2 similarly improve upon their forerunners.
A 6% improvement over the mark delivered by the MaXLine III propels the MaXLine Pro to the top of the charts in our Sims 2 disk trace. WD's newer drive, on the other hand, stumbles again when contrasted with the RE2 400. Seagate ekes out a slight improvement towards the bottom of the chart.
When it comes to the ever-popular World of Warcraft, WD's newest once again tops the charts at 592 I/Os per second. The NL35.2 claws its way back into the middle of the chart with a respective 535 IOps while the MaXLine Pro finds itself at the bottom with 503 IOps.
Unlike single-user machines (whether a desktop or workstation), servers undergo highly random, non-localized access. StorageReview simulates these multi-user loads using IOMeter. The IOMeter File Server pattern balances a majority of reads and minority of writes spanning requests of varying sizes.
IOMeter also facilitates user-configurable load levels by maintaining queue levels (outstanding I/Os) of a specified depth. Our tests start with the File Server pattern with a depth of 1 and double continuously until depth reaches 128 outstanding I/Os.
Drives with any sort of command queuing abilities will always be tested with such features enabled. Unlike single-user patterns, multi-user loads always benefit when requests are reordered for more efficient retrieval.
For more information click here.
Though they feature a SATA interface, multi-user/server workloads nonetheless remain the raison-d'etre of these enterprise-class drives. Under any kind of concurrency, WD's 500 GB RE2 enjoys a consistent lead over the competition. The newer unit builds nicely upon the older 400 GB drive's scores, marks that already led the pack when it came to 7200 RPM drives.
The MaXLine Pro delivers slightly better performance than its predecessor under lower (read practical) queue depths but slides a bit when concurrency hits 16 or greater. Seagate's newer drive delivers more or less the same performance as the original NL35 but also slips a bit under heavy loads.
Noise and Power Measurements
Idle Noise- The sound pressure emitted from a drive measured at a distance of 3 millimeters. The close-field measurement allows for increased resolution between drive sound pressures and eliminates interactions from outside environmental noise. Note that while the measurement is an A-weighted decibel score that weighs frequencies in proportion to human ear sensitivity, a low score does not necessarily predict whether or not a drive will exhibit a high-pitch whine that some may find intrusive. Conversely, a high score does not necessarily indicate that the drive exhibits an intrusive noise profile.
Operating Power Dissipation- The power consumed by a drive, measured both while idle and when performing fully random seeks. In the relatively closed environment of a computer case, power dissipation correlates highly with drive temperature. The greater a drive's power draw, the more significant its effect on the chassis' internal temperature.
Startup (Peak) Power Dissipation- The maximum power dissipated by a drive upon initial spin-up. This figure is relevant when a system features a large number of drives. Though most controllers feature logic that can stagger the spin-up of individual drives, peak power dissipation may nonetheless be of concern in very large arrays or in cases where a staggered start is not feasible. Generally speaking, drives hit peak power draw at different times on the 5V and 12V rails. The 12V peak usually occurs in the midst of initial spin-up. The 5V rail, however, usually hits maximum upon actuator initialization.
For more information, please click here.
WD's 500 GB RE2 trims the noise exhibited by its predecessor by a considerable margin. At just 39.5 A-weighted decibels when idle, the drive comes in among the quietest we have ever objectively measured. Seagate's NL35.2 also improves upon the firm's older unit. Only Maxtor regresses slightly, with the newer MaXLine Pro reaching a relatively high 44.1 d/BA.
Subjectively speaking, seeks for the NL35.2 and 500 GB RE2 share a similar dull, muted character. The MaXLine Pro's seeks are just slightly more substantive than the other two drives. All are, of course, several notches quieter than the typical 10,000 RPM SCSI disc.
Of the 500-gigabyte units, WD's drive delivers the lowest operating power dissipation at 8.6 watts when idle. Such a score weighs in slightly higher than the drive's predecessor (8.4 watts) as well as Seagate's original NL35.1. The MaXLine Pro and Nl35.2 weigh in at 9.0 watts and 9.4 watts respective.
When undergoing 100% seek activity, however, margins widen. Here the 500 GB RE2 shines with a power draw of just 10.6 watts, the thriftiest draw of any 400- or 500-gig unit. The NL35.2 comes in at 13.4 watts, drawing significantly more power than its predecessor. The MaXLine Pro bottoms out the chart through dissipating 14.5 watts.
All newer 500 GB units draw significantly more power when spinning up from a powered-down state. Once again, of the larger discs, WD's 500 GB drive rests at the top. The RE2 500 can not, however, quite match the score turned in by its predecessor nor that delivered by the older, three-platter MaXLine III. Maxtor's newer MaXLine Pro, however, draws a full 10 watts more than the III while the NL35.2 brings up the rear with a hefty spin-up current draw.
How does NCQ affect performance?
StorageReview has found that, generally speaking, past NCQ implementations have exacted a slight across-the-board performance penalty for single-user applications. A notable exception was Maxtor's MaXLine III. One of the first drives to feature NCQ, the MaXLine III tended to perform slightly better with NCQ enabled.
These trends carry through into the current crop of 500-gigabyte units. While the Seagate NL35.2 and WD RE2 500 suffer small losses in single-user assessments with NCQ enabled, the Maxtor MaXLine Pro enjoys minor increases.
As always, it is in the multi-user arena that command reordering truly shines. With NCQ disabled, all three drives fail to exhibit any increase in IOps whatsoever as concurrency increases.
Featuring top performance, the friendliest environmental measurements, and lowest price, WD's RE2 WD5000YS makes a compelling case as the drive of choice when it comes to capacious 500 GB SATA drives intended to serve enterprise needs that do not require the utmost random access speeds.
The WD5000YS builds upon the leading performance delivered by its predecessor while managing to significantly trim both acoustic output and power draw when undergoing activity. The manufacturer argues that the latter point is especially important; a drive that draws less electricity saves on the power bill, requires less cooling, and undergoes less stress, presumably increasing drive longevity.
While it offers adequate multi-user performance, the Seagate NL35.2 commands a hefty price tag and as a result can not compete against WD's offering when it comes to IOps per dollar. The drive does, however, enjoy Seagate's name as well as a great track record if recent SATA Barracuda drives are any indication.
Maxtor's MaXLine Pro finally builds upon the aging 300 GB design of the original MaXLine III. From a performance perspective, however, the Pro offers about the same performance as the III while introducing more noise and drawing more power.