Big, Fast IDE Drives

The first high-capacity IDE drives show a lot of promise, particularly if you have a local-bus system that supports them well

David Essex

The need for fast, high-capacity hard drives has recently challenged the well-established popularity of the ATA (AT Attachment) interface used with IDE drives. The low cost and relative technical simplicity of IDE drives are little help when the data-storage requirements of workstations and multimedia machines push against a BIOS- and DOS-imposed 528-MB capacity limit. Further, on local-bus systems, ATA's 2- to 3-MBps data-throughput rate is no match for the nominal 5- to 20-MBps rates of the SCSI bus, putting IDE at a speed disadvantage in multimedia environments.

To meet this challenge, last year hardware and system software makers began updating the ATA standard--push ed somewhat by an Enhanced IDE standard proposed by Western Digital (see ``IDE Takes Off,'' March BYTE). Parts of Western Digital's proposal relating to increased capacity (well beyond 528 MB) and higher ATA bus throughput were adapted by the SFF (Small Form Factor Committee), an ad hoc industry group. The resulting SFF proposal is now in the hands of ANSI, where it could pick up other enhancements, such as CD-ROM support, on its way to becoming known as the ATA-2 standard.

A few (540-MB) drives began implementing the SFF guidelines last year. This past spring and summer, larger, faster drives arrived, and they are the subject of this roundup. All offer burst data transfer rates over the ATA interface of at least 11.1 MBps. The performance story remains more complicated, however, as the data density of magnetic media, spin rates, and efficiency of read/write head movement continue to play major roles. The point is well illustrated by the varying throughput and seek speeds exhibited by the reviewed driv es, which range in size from 540 MB to 1.08 GB.

In the Channel

The eight IDE drives reviewed here are representative of hard drive manufacturers' latest efforts to bring drive electronics up to speed with local-bus systems and improvements in the mechanical aspects of drive architecture. Standard ATA data transfers going over the typical ISA bus are limited to a bandwidth that cannot handle the faster stream of bits created by newer drives. Growing ever smaller, these drives have more densely packed platters with drive heads that are themselves getting smaller, lighter, and quicker, both mechanically and electronically. Even when matched with higher-throughput local-bus computers, IDE drives could not match SCSI rates, because the CPU's standard way of controlling throughput (called Processor I/O, or PIO) did not let it recognize when 100 percent of the shared local bus's bandwidth was available for data transfers.

The solution to faster PIO, implemented on all the reviewed drives, is call ed flow control using IORDY (the I/O Channel Ready line on the ATA bus). Flow control lets the drive electronics regulate the CPU's data transfer functions so the drive can reliably communicate with the bus at higher speeds. The SFF has standardized new ATA timings using flow control (called PIO Mode 3) at 180 nanoseconds per cycle. The result, in theory, is maximum internal transfer rates, or burst rates, of up to 11.1 MBps. That translates into at least a doubling of effective data transfer rates, a result more or less borne out by the BYTE benchmark tests. I measured sustained buffer-to-host rates as high as 7 MBps.

An alternative method, Fast Multiword DMA, uses the computer's DMA controller instead of the CPU to handle data transfers between the drive and system memory. SFF defines Mode 1 DMA transfers to allow burst rates of up to 13.3 MBps, roughly equivalent to PIO Mode 3. (A new PIO Mode 4 offers burst rates of up to 16.6 MBps and shows up here only on the 1-GB IBM 0662 Model A10 and Seagate S T31220A, although systems weren't available at review time to support or test these rates.)

Faster throughput was not the only IDE improvement I hoped to observe in the tested drives. The SFF proposal provides two translation schemes that let DOS-based systems break the 528-MB barrier. The first, CHS, involves a simple translation between the slightly incongruous cylinder-head-sector parameters in the system BIOS and those in the drive firmware.

The second scheme, LBA (logical block address), translates the CHS information into a 28-bit address that can be used by the operating system, device driver, and drive interface. All the reviewed drives support the more advanced LBA standard, but they use the simpler CHS in older systems with BIOS code that doesn't support LBA. Many newer systems also support only CHS, because it's more difficult to modify a BIOS to support LBA. The CHS scheme has an upper capacity limit of 8.4 GB--enough for now.

Small size and energy efficiency are two more bene fits common to these drives. None is more than 1 inch high; Seagate's 540-MB ST5660A is only 0.75 inch high (it's also smaller than the others in length and width). These reduced dimensions will allow computer manufacturers to continue downsizing system boxes. In addition, the drives all boast low wattages, helping to meet federal government procurement requirements specified in the Environmental Protection Agency's Energy Star program.

Drive Testing

I used two PC systems to test the hard drives: a Compaq Deskpro XE 450 (50-MHz 486DX2) and a Zeos Pantera (60-MHz Pentium). Each system provided a built-in local-bus IDE connection and a BIOS that supports IDE drives larger than 528 MB. The Compaq was a year-old ISA system with a 170-MB Conner IDE drive; the Zeos was a more recent PCI (Peripheral Component Interconnect) design with a 425-MB Western Digital IDE drive. In both systems, the host drive was configured as an IDE master and the test drives as slaves, using the appropriate drive jumpers if n ecessary.

I used three tests: the file I/O parts of BYTE's low-level DOS benchmark, Quantum's QBench 1.30, and NSTL's PLATT (Page Level Availability Time Test) (see ``Lab Report: 32 High-Speed Hard Drives,'' September 1993 BYTE). Each test takes its own read on the complexities of drive performance.

The BYTE file I/O test measures read and write I/O from the DOS file level. QBench works at the BIOS level, measuring sequential and random reads and writes. It produces a weighted average of access times and data transfer rates based on profiled DOS applications (65 percent sequential I/O, 35 percent random; 60 percent read, 40 percent write).

The NSTL PLATT benchmark is based on a profile of hard drive activity measured from Windows business applications running with Microsoft's SMARTDRV caching program. The access-time results from 12 tests, a mix of random and sequential read and write operations, are weighted and averaged according to the Windows profile. One part of the PLATT allows cal culation of sequential throughput (dividing the 16-KB block size by the returned access time in milliseconds).

The age difference between the two test systems influenced drive installation and performance. While the Compaq automatically worked with the two Western Digital drives and the 1-GB Seagate, it required utility software such as OnTrack's Disk Manager (provided with many of the drives) to recognize the other drives, and it wouldn't work with the 540-MB Seagate unit unless set for 528 MB. The BIOS in the newer Zeos system recognized and worked with all tested drives, automatically configuring to their maximum capacity through a setup menu selection.

The Zeos, with its more sophisticated IDE controller, got much higher throughput rates in many of the tests, showing that the system can be a bottleneck with new, faster drives. Testing the five biggest drives on both systems showed that the 1-GB IBM 0662 Model A10 picked up the greatest throughput benefit from the faster system; the Micropoli s Taurus 4110A, the 728-MB IBM DSAA-3720, and the Western Digital Caviar AC31000 showed intermediate gains. The 1-GB Seagate drive made smaller but still substantial throughput gains. The fastest measured throughput, 7 MBps, came from the 1-GB IBM drive on a PLATT sequential-read test that pulls most of its data from the drive's built-in buffer. The test system's ATA bus speed had no effect on access-time tests.

IBM DSAA-3720 and 0662 Model A10

Although the two IBM drives differ from each other substantially, both employ the magnetoresistive head technology invented by IBM. Magnetoresistive read and write heads require fewer copper coils and material layers than standard thin-film heads. The benefit, according to IBM, is smaller, simpler heads that allow higher data density and better durability. Both IBM drives also employ digital read channel electronics that allow reliable reading of more densely packed data (see ``Digital Hard Drives,'' March BYTE).

These two IBM drives do, in fact, ha ve some of the highest track density of the group (as high as 4300 tracks per inch for the 728-MB DSAA-3720), and the 1-GB 0622 claims a top-rated MTBF (mean time before failure) rating of 500,000 hours.

The 1-GB IBM drive's smaller, more efficient heads and higher data density do not seem to give it any overall performance advantage over the 1-GB Micropolis Taurus, however. The Micropolis makes up in access time what it loses in data transfer rate. In the file I/O test, the Micropolis drive bested the field and exceeded the 1-GB IBM unit's speeds by 3 percent to 13 percent. The IBM drive fared better in the NSTL PLATT benchmark, where its higher throughput slightly edged out the Micropolis drive's quicker access times.

The IBM unit did slow down in the PLATT sequential test that accesses four areas of the drive simultaneously (an indication of server capability), but it was still as fast as or faster than any other drive but the Micropolis Taurus in this test. Both 1-GB drives have a large (512 -KB) buffer, so differences in caching algorithms are the most likely reason for the IBM drive's performance drop-off.

IBM's 728-MB DSAA-3720 has the smallest cache in the group (96 KB), and, at 4500 rpm, its rotation is slower than the 1-GB drives'. But it compares favorably to the 1-GB IBM drive for multimedia uses, because it doesn't have to perform TCAL (thermal recalibration), a periodic compensation for heat-induced mechanical changes that can translate into lost video frames. (Some of the other reviewed drives don't need to do TCAL, either.) IBM says it's preparing a workaround to mitigate TCAL's effects on the 1-GB drive.

The small cache may be one reason why the 728-MB drive has lower throughput than two of the smaller (540-MB) drives when, theoretically, its higher track and areal density would let it pump bits faster. In the file I/O tests, the 728-MB IBM drive generally falls slightly behind the 540-MB Quantum and Seagate units, although it passes two 1-GB drives on the Pentium QBenc h test.

Micropolis Taurus 4110A

As noted above, the 1-GB Micropolis Taurus performs slightly better than the 1-GB IBM drive even though it lacks the data density advantages of magnetoresistive technology. It avoids some of the downside of high capacity by employing five platters. It thus has two more than the three-disk IBM drive, giving it an edge when performing nonsequential seeks on large files (measured by the Short Seek and Long Seek tests in the PLATT benchmark).

The Micropolis can confine such files to a narrower range of cylinders (tracks that are the same general distance from the drive spindle) because the extra disks allow more capacity per cylinder. In other words, the more stacked up the cylinders, the shorter the distances the read/write heads must move when randomly accessing in a file.

Both IBM and Micropolis state that the 1-GB drives' 5400-rpm rotational speeds reduce latency time (the time it takes for the requested block to pass under the head once the head has arrived at the correct track) to 5.56 milliseconds. This is an easily proved mathematical verity and, coupled with the low average seek time (8.5 ms) claimed for both drives, should lead to quick data access. In those tests measuring random-access times in the Windows-oriented PLATT benchmark, the two drives did better than all the others, with the Micropolis generally faster than the IBM. In the DOS-oriented QBench test, however, the two Western Digital drives were fastest in data-access-time tests.

Using more disks contributes to the Micropolis's high $739 street price, but then Micropolis offers an industry-leading five-year warranty. (IBM offers a five-year warranty on its 1-GB IDE drive but not on the 728-MB unit.) Those looking to push the capacity limits of the IDE interface ought to seriously consider this drive.

Quantum ProDrive LPS 540AT

While the ProDrive LPS 540AT is larger than 528 MB and supports fast local-bus transfer modes, it won't be Quantum's fastest or largest IDE driv e by the time you read this. Quantum's new 540-MB Lightning drive should be faster, and a 730-MB version of the Lightning will be larger. While relatively slow in the group reviewed here, the ProDrive LPS 540AT is still a fast drive with an attractive price. The drive supports PIO Mode 3 and Fast Multiword DMA, so the ATA interface throughput can potentially match the drive.

However, in both the access-time measures important for random file I/O and the throughput measures important for sequential file I/O, the drive placed well back in the pack (though rarely last). Among the three 540-MB drives reviewed here, it placed second in sequential-access tests, behind the Seagate ST5660A. The ProDrive fared best in several of the simpler PLATT sequential tests, indicating that it may be a better drive for Windows than DOS, since that test is modeled on the use of hard drives in a Windows environment.

Two factors may have depressed the throughput results. Like the Western Digital drives, the Quantum Pr oDrive has a fairly small (128-KB) cache/buffer. And at 2875 tpi, it has the second-lowest track density of the group (the Micropolis Taurus 4110A has the lowest).

One final note: Quantum, like Western Digital and Micropolis, ships its drive with a specially licensed copy of OnTrack Computer Systems' Disk Manager to perform the CHS translation needed to get above the 528-MB DOS/BIOS barrier on systems with older BIOS code.

Seagate ST5660A and ST31220A

Seagate's mid-capacity entry has little that overtly differentiates it from the other 540-MB drives. When tested, however, the ST5660A showed seek and throughput speeds that put it not only at the top of the 540-MB group but in range of the 1-GB drives. For example, in the file I/O read test, the ST5660A's 559.4 KBps placed it fourth overall, behind three 1-GB drives. In the PLATT benchmarks, it jumped toward the top in the sequential tests.

Note that the Seagate was the only drive I couldn't configure for more than 528 MB on the Comp aq test system. The Compaq's BIOS didn't automatically recognize it, and Seagate has no plans to offer its own or a third-party driver. That shouldn't be a problem later on in the year, as the Phoenix BIOS, AMI's latest BIOS, and others designed to handle higher IDE capacities and LBA translation become more widely available. However, if you own an older system, you may not be able to access the ST5660A's top 12 MB without a utility like OnTrack Disk Manager.

The Compaq had no trouble with Seagate's 1-GB ST31220A Medalist 1080. Formatting nicely at over 1079 MB, the Medalist performed competitively with the 1-GB IBM and Micropolis drives on the Compaq, running neck-and-neck with Seagate's smaller 528-MB unit. Both Seagate drives did well on sequential tests, but not as well with random I/O. Read results were consistently behind writes for both Seagates (they use the same caching algorithms). On the Zeos Pentium system, the 1-GB Seagate fell significantly behind other 1-GB drives on throughput tests.

Western Digital Caviar AC2540 and AC31000

Although Western Digital, champion of the Enhanced IDE movement, has endowed its Caviar IDE drives with all the right high-speed protocol support--Mode 3 PIO and Mode 1 DMA--its efforts do not seem to have brought superior performance. The drives are fast, but not as fast as the others I tested. The 540-MB AC2540 and 1-GB AC31000 provided similar results in most tests, with the larger, newer drive coming out a bit ahead.

The Western Digital drives placed last or next-to-last in most throughput tests. For example, the AC2540's 495-KBps result placed it last in the file I/O read test. In the PLATT benchmark, the Western Digital drives placed in the slower group with the 540-MB Quantum and 728-MB IBM drives. QBench results were mixed: slowest in transfer time but fastest in access time. The QBench access-time result derives from both sequential and random operations, so the built-in cache helps out. The PLATT benchmark's seek tests, where the Caviars di dn't do as well, are based on random I/O and more reflect the drives' mechanical abilities.

But keep performance in perspective. The Western Digital drives offer capacities and seek times that would have been unheard of in an IDE drive two years ago. If you have an older ISA system that you're upgrading for increased capacity, look at Western Digital's pricing, because you aren't likely to see great performance differences between drives without local bus.

With capacities over 1 GB and support for local-bus transfer rates, new IDE drives remain your best bet for desktop systems, where lower drive-interface overhead gives them a speed advantage over comparable SCSI drives. SCSI shines only in multitasking, multidrive situations and can provide multigigabyte capacities. Drive utility programs like OnTrack's let you use these drives on just about any system with IDE connectors, but newer systems that automatically recognize large capacities and provide local-bus throughput are where they belong.

---

The Facts

IBM Storage Systems Division
(DSAA-3720; 0622 Model A10)
Dept. 2W4
Rte. 100, P.O. Box 100
Somers, NY 10589
(800) 426-7299
fax: (800) 426-3395


Micropolis
(Taurus 4110A)
21211 Nordhoff St.
Chatsworth, CA 91311
(800) 395-3748
(818) 709-3300
fax: (818) 709-3396


Quantum Corp.
(ProDrive LPS 540AT)
500 McCarthy Blvd.
Milpitas, CA 95035
(800) 624-5545
(408) 894-4000
fax: (408) 894-3205


Seagate Technology
(ST5660A; ST31220A Medalist 1080)
920 Disc Dr.
Scotts Valley, CA 95066
(408) 438-6550
fax: (408) 438-7852


Western Digital Corp.
(Caviar AC2540; Caviar AC31000)
8105 Irvine Center Dr.
Irvine, CA 92718
(800) 832-4778
(714) 932-5000
fax: (714) 932-6498

---

IDE Hard Drive Features

                                                NO. OF
                               CAPACITY  DISKS  HEADS/      SPIN    AVERAGE
                               (MB)             RECORDING   RATE    LATENCY
DRIVE
                                           SURFACES    (RPM)   (MS)
IBM DSAA-3720                  728       2      4/4         4500    6.67
IBM 0662 Model A10             1052      3      6/5         5400    5.56
Micropolis Taurus 4110A        1057      5      10/9        5400    5.56
Quantum ProDrive LPS 540AT     541       2      4/4         4500    6.67
Seagate ST5660A                546       2      4/4         4500    6.67
Seagate ST31220A Medalist 1080 1083      3      6/6         4500    6.67
Western Digital Caviar AC2540  541       2      4/4         4500    6.67
Western Digital Caviar AC31000 1084      3      6/6         4495    6.67


                                SEEK TIMES (MINIMUM/       CACHE/     TRACK
                                AVERAGE/MAXIMUM; MS)*      BUFFER     DENSITY
DRIVE                                                      SIZE (KB)  (TPI)
IBM DSAA-3720                   4/12/25                    96         4300
IBM 0662 Model A10              0.6-2.5/8.5-10.1/16.
5-18   512        4077
Micropolis Taurus 4110A         1.5/8.5/20                 512        2756
Quantum ProDrive LPS 540AT      4/12/23                    128        2875
Seagate ST5660A                 3.5/12/30                  256        3309
Seagate ST31220A Medalist 1080  3.5/12/25                  256        4250
Western Digital Caviar AC2540   4/11-13/25                 128        3300
Western Digital Caviar AC31000  4/10-12/23                 128        4000


                                    RATED MTBF   WARRANTY   PRICE
                                    (POWER-ON    (YEARS)
DRIVE                                HOURS)
IBM DSAA-3720                       300,000         2       $435 (list)
IBM 0662 Model A10                  500,000         5       $730 (list)
Micropolis Taurus 4110A             500,000         5       $739 (street)
Quantum ProDrive LPS 540AT          300,000         2       $389 (street)
Seagate ST5660A                     300,000         2       $379 (street
)
Seagate ST31220A Medalist 1080      300,000         3       $689 (street)
Western Digital Caviar AC2540       250,000         3       $349 (street)
Western Digital Caviar AC31000      250,000         3       $679 (street)


Specs like these can give a good picture of a drive's capabilities, although they do not take into account one important attribute: the effectiveness of the caching algorithms for the drive's built-in cache. Larger drives get the best technology--for example, the higher (5400-rpm) spin rates for the 1-GB IBM and Micropolis drives. They also have higher MTBF ratings and the longest warranties.


* Track-to-track, including settling time; hyphen separates read and write values where necessary

---

Illustration: Graph: Drive Performance Results Results from three different hard drive tests show a fairly consistent performance picture, with slight jostling of relative positions between tests. Bigger drives generally do better because they use the latest technology .

---

Illustration: Graph: On a Faster System Running on a 60-MHz Zeos Pentium system with a more sophisticated IDE interface than the 50-MHz Compaq 486 Deskpro had, several of the drives improved markedly in throughput results, as shown here by the QBench test. The 1-GB IBM and Micropolis drives showed the strongest benefit.

---

Photograph: The new generation of fast IDE drives with larger capacities is a good match for fast local-bus systems. Counterclockwise from left are the 1-GB Micropolis Taurus 4110A, the 1-GB IBM 0662 Model A10, the 540-MB Seagate ST5660A, the Western Digital Caviar AC2540, the 728-MB IBM DSAA-3720, and the 540-MB Quantum ProDrive LPS 540AT. Not included in the photo are Seagate's 1-GB ST31220A Medalist 1080 and Western Digital's 1-GB Caviar AC31000.

---

David Essex is a BYTE technical editor. You can reach him on the Internet or BIX at dessex@ bix.com.