Design Features

Disk drives at 40
Lean, mean storage machines
Maury Wright, Technical Editor

The disk drive may have turned 40 this year, but you won't find symptoms of middle age creeping up on this all-star performer. New drives stretch to 9-Gbyte storage capacities and achieve data-transfer rates approaching 150 Mbps. Meanwhile, ubiquitous 2-Gbyte drives for PCs sell for $200 at discount stores.

  System designers who need disk drives for projects ranging from compact embedded systems to multiprocessor enterprise servers need to take a close look at the latest offerings from drive vendors. Sure, the drives store more and cost less than they did last year, but new products feature entirely new interfaces, form factors, and recording technologies, of which designers need to be aware. Moreover, the breadth of offerings, which ranges from 1.8-in. drives that compete with flash memory for PC Card applications to 9-Gbyte, 3.5-in. drives, can be confusing.

  Most designers face a two-stage disk-drive evaluation. First, you must prioritize some general characteristics of disk drives for your application. For example, a designer of notebook computers might rate capacity as the most important characteristic followed in order by performance, ruggedness for the mobile environment, power consumption, and price. In contrast, a designer working on an enterprise server might establish performance, reliability, availability, future-product road map, and price as an appropriate pecking order.

  In the second stage of evaluation, the designer must map features of different disk drives to the prioritized list of general requirements. Some factors, such as capacity, map directly from a drive spec sheet to the prioritized list. Conversely, a characteristic such as performance depends on a mix of interrelated disk-drive specs, including seek time, rotation speed, media-transfer rate, buffer size, interface type, and interface-transfer rate.

  Table 1 provides a broad summary of today's disk-drive market broken down by application or market segment. The cost per megabyte should provide a good indication of the OEM cost of drives. The low end of the ranges listed typically coincides with large volume purchases of the highest capacity drives available. Conversely, the high end of the price range is indicative of low-volume purchases of lower capacity drives.

Heads and the read channel

  Perhaps you should start any investigation of disk drives with a look at the internal head and read-channel technologies that writers have prominently featured in recent storage articles. Just as EDN predicted 18 months ago (Reference 1), the industry has begun to ship PRML (partial-response maximum-likelihood) read channels and MR (magnetoresistive) heads in leading-edge drives.

  Basically, MR heads include an MR-read element that can read more closely spaced magnetic flux transitions than can earlier thin-film inductive heads. MR heads also include a separate inductive write element optimized for producing closely spaced flux transitions. PRML channels, however, use signal-processing techniques to map signal shapes read from a disk to expected signal shapes. One analogy compares PRML channels to finding mountains of data with a map, whereas previous-generation peak-detect channels assumed a mountain of data at each signal peak. Actually, PRML channels work similarly to data-communication products such as modems in decoding useful data in the presence of noise.

  Without question, PRML channels and MR heads have allowed significant increases in capacity and peak media data rate. PRML alone can provide a 30% improvement in areal density compared with peak-detect channels. Note that manufacturers do not necessarily need to use both MR heads and PRML channels.

  In some cases, system designers don't need to know even what type of channel or head a disk uses. These technologies certainly influence capacity and performance specs, as well as cost, but, at first glance, the technologies don't have significance beyond these specs. Designers should be aware of a few market and technological facts, however, regarding read channels and heads.

  One such fact is that MR heads are still in the early stages of volume usage. Dennis Waid, president of Peripheral Research Corp (Santa Barbara, CA), claims that for calendar year 1996 the drive industry will ship 175 million MR heads of a total of 700 million heads shipped. Moreover, more than 60% of the MR heads shipped are manufactured by IBM for IBM drives, and only IBM has achieved manufacturing yields on MR heads that truly make the technology economically feasible. According to Waid, IBM's yield will exceed 62% in 1996, and the yields for other MR-head manufacturers will remain 15 to 28%.

  System designers can conclude that IBM is able to produce drives based on MR heads at a reasonable price and without limits on availability. You would expect to pay a significant price premium for MR-based drives from other companies. In fact, drive vendors currently pay $7.50 to $10 for thin-film heads. According to Waid, MR heads would currently cost approximately $25 to $30 if head vendors based the price on the actual manufacturing cost. Instead, the head manufacturers are now selling MR heads for $15 to $16, and Waid expects prices to possibly drop to $10 next year. The head manufacturers are intentionally losing money now to ramp production and gain market share looking toward 1998 when virtually every drive will use MR heads. With the current state of the MR-head market, designers might be wary of committing to an MR-head drive if availability and price are paramount to an application. On the other hand, designers who need the highest capacity and fastest transfer rates may have no other choice.

  PRML read channels, conversely, have matured significantly, and a number of companies, including Lucent Technologies (Allentown, PA), Silicon Systems (Tustin, CA), Marvell Semiconductor (Cupertino, CA), and Cirrus Logic (Fremont, CA), are producing the channels in volume. PRML channels still may cost slightly more than peak-detect channels but generally result in higher capacity, faster data rates, and even lower error rates. PRML does have one vulnerability for designers working in power-critical applications, such as notebook computers: PRML channels generally require almost 1W of power, although peak-detect channels use less than 0.5W. In fact, even IBM has stuck with a peak-detect channel in all of its 2.5-in. products except the newest 2.1-Gbyte member of its Travelstar family.

  Regardless of your opinion on heads and read channels, you can buy your technology of choice in a variety of form factors. As Table 1 indicates, some of the form factors have direct relationships to specific applications and market segments, whereas the most prevalent 3.5-in. drives cross a number of market boundaries.

  Starting at the smallest form factor, the market for 1.8-in. drives has slipped from a handful of manufacturers just two years ago to only Intégral today. Analysts attribute the lack of interest from major drive vendors to the failure of products such as personal digital assistants (PDAs) and palmtop PCs to catch on as mainstream products. Most of the vendors and analysts, however, believe that the PDA market will eventually explode, so don't think players such as Seagate and IBM have decided to forgo the 1.8-in. market permanently.

  Meanwhile, Intégral offers a range of products in its Viper family. According to Intégral, the bulk of its sales comes in the 170-Mbyte capacity, unlike other market segments in which manufacturers have discontinued such capacities. The PC Card-sized drives, however, target specialized applications in embedded systems and high-capacity removable storage for notebook computers. In fact, Intégral claims that it competes against flash memory more than against other disk drives. Intégral has announced a 510-Mbyte drive but has supplied only sample quantities at that capacity.

2 Gbytes in a 2.5-in. package

  In the market for 2.5-in. drives, vendors are attempting to track the capacity ramp of 3.5-in. products. Manufacturers believe that they must closely follow 3.5-in. capacity to serve a notebook market in which customers increasingly demand a feature set that matches that of high-end desktop computers. As you might guess, 2.5-in. products use more aggressive technology than do commodity 3.5-in. drives to achieve maximum capacities that now exceed 2 Gbytes. This use of technology combined with lower volumes results in the substantial price premium vendors charge for 2.5-in. drives.

  Currently, IBM and Toshiba hold dominant positions supplying 2.5-in. drives. It's probably no coincidence that these two companies are the top suppliers of notebook computers and captively use many of the 2.5-in. drives they produce. IBM is currently shipping its three-platter Travelstar 2XP drive, which stores 2.16 Gbytes in a 17-mm-high package. The company also offers 1- and 1.44-Gbyte family members that measure only 12.5-mm high and include two platters.

  Toshiba is shipping its 2.16-Gbyte MK-2101 family, which measures 19 mm high and uses five platters, a PRML channel, and thin-film heads. Around press time, however, the company will begin shipping the HDD-2716, which provides the same capacity using MR heads with three platters. Toshiba has even been able to package the three platters in a 12.5-mm, low-profile design.

3-in. platters double surface

  With 2.5-in. drives established as a notebook standard, you might think that no other form factors are important to the market segment. Certainly, smaller form factors, such as 1.8-in. drives, have proven unsuitable for usage in mainstream notebooks. JTS, however, believes that notebooks could more effectively use a slightly larger drive. JTS also believes that its 3-in. Nordic Plus family can solve the conflicting requirements of high capacity and low cost in a small drive.

  According to JTS, its 3-in.-platter drives offer 80% more recording surface than do typical 2.5-in. drives. JTS believes that building a thin drive will ultimately be more important to notebook manufacturers than will the form factor or footprint. JTS is taking advantage of the larger recording surface to use mature and low-cost head and read-channel technology. In fact, the Nordic family uses MIG (metal-in-gap) heads that were widely used before thin-film heads. JTS now offers a 1-Gbyte drive, the most popular capacity in notebook computers, that measures only 10.5 mm high and uses two platters. The drive features an areal density of 410 Mbps/in.₂, and 1-Gbyte, 2.5-in. drives must store more than 900 Mbps/in.₂. Perhaps even more impressive is a two-year road map. JTS claims it can offer 3-Gbyte drives in the second half of 1997 with an areal density of only 1000 Mbps/in.₂. The technologies required to achieve a 1000-Mbps/in.₂ density will certainly be mainstream at that point. JTS claims that the 3-in. form factor will allow it to maintain a 30 to 40% advantage in cost per megabyte relative to vendors of 2.5-in. drives.

  At first glance, the 3-in. product seems to have no disadvantages, but vendors of 2.5-in. drives are quick to point out some trouble spots. For example, they claim that the larger platters require more power for spindle motors, yet the target applications are power-sensitive. In addition, manufacturers need to ruggedize notebook drives to withstand significant nonoperating shock. The vendors of 2.5-in. drives also claim that the increased mass of 3-in. products makes it significantly harder to achieve a rugged design. Finally, they point out that the material cost is higher for the larger products, and that, although the 3-in. drives use standard heads, they require nonstandard media that manufacturers do not currently produce in high volume.

  Predicting the success of 3-in. products really comes down to whether a significant number of notebook vendors accommodate the drives in new designs. JTS claims that, excepting notebooks from Toshiba and IBM, most new systems will accommodate either 2.5- or 3-in. drives in the near future, but Toshiba and IBM claim that virtually no systems will accommodate the larger drives in the near future. Phil Devin, vice president of analysis company Dataquest (San Jose, CA), believes that substantial slots will be available for 3-in. drives, and Dataquest issued a positive perspective on the form factor this June.

  One other factor that could boost the chances of the 3-in. form factor is support from other drive vendors. Western Digital, in fact, has licensed the technology from JTS, and, although Western Digital has not introduced a 3-in. product yet, it may do so at Comdex in November in Las Vegas. Dataquest's Devin also claims that several other drive vendors, including NEC, Samsung, and Intégral, are considering 3-in. products. Should the technology succeed in mainstream computing markets, 3-in. products could quickly become popular choices for designers of embedded systems who could take advantage of the low cost and relative smallness.

5.25-in. drives threaten a return

  Just as JTS pitches the 3-in. drive as a value proposition, Quantum has made a similar proposition for returning 5.25-in. drives to the desktop. In February, the company announced its Bigfoot family with drives offering 1.2- and 2.5-Gbyte capacities. Quantum believes that the CD-ROM has re-established the 5.25-in. slot in desktop PCs and that most desktop enclosures will include a mounting slot capable of hosting a Bigfoot drive. In fact, standard CD-ROM bays measure 15/8 in. high, and you can almost squeeze a Bigfoot and a low-profile CD-ROM into a standard CD-ROM bay.

  Quantum reasons that building a Bigfoot drive costs slightly more than building a 3.5-in. drive, but a Bigfoot with one platter is certainly less expensive to build than a 3.5-in. drive with two platters. Because the Bigfoot can store almost as much on one platter as a 3.5-in. drive stores on two platters, Quantum claims that the larger form factor can offer cost savings of $0.02 per megabyte, a substantial amount in the commodity-PC market.

  Quantum designed the Bigfoot using the same technology base of its mainstream Fireball 3.5-in. drives, including thin-film heads and a PRML channel. Although the Fireball family rotates at 4500 rpm, the Bigfoot spins its platters at 3600 rpm. The reduction in rotational latency was necessary because of the larger diameter of the outer tracks on the disks. Had the drive spun at 4500 rpm, the data rate from those outer tracks would have exceeded the capabilities of the heads and read channel.

  Drives with higher spin rates generally offer better performance because the higher rate generates higher linear bit density and, therefore, faster data-transfer rates. Also, higher rotational speed minimizes the rotational latency—the time it takes the disk to rotate to the sector of interest once the heads are aligned on track. Generally, manufacturers specify rotational latency as the time required for one-half of a rotation. The larger disk in Bigfoot also contributes to a 15-msec average seek time compared with mainstream 3.5-in. products that feature 12-msec or faster seek times. Competing vendors have been quick to highlight the low rotational speed and seek time of Bigfoot.

  Performance, however, is a multifaceted characteristic. For example, the longer outer tracks Bigfoot offers in some cases allow for lengthier and faster sustained data transfers than could a 3.5-in. drive rotating at a faster speed. The advantage is due to the need for 3.5-in. drives to mechanically reposition the head to a different track more often. It's unclear whether a typical PC user could discern the difference between a Bigfoot and Quantum's standard Fireball drive. Still, competitors have forced Quantum to position the Bigfoot as a value to users more interested in capacity than performance.

  In general, industry observers do not expect major success for Bigfoot. Dataquest's Devin points out two problems for Bigfoot from the OEM perspective. First, Devin claims that even when 5.25-in. slots are available in a PC case, the OEM will want to reserve the "money slots" for high-margin peripherals, such as tape drives or optical drives. Second, the lengthier outer tracks will continue to limit the performance achievable with mainstream head and read-channel technologies.

  At one time, a number of other companies, including Seagate and Maxtor, were believed to be pursuing 5.25-in. designs. Today, in addition to Quantum, only JTS has such products on its published road map. Still, Quantum plans a second-generation Bigfoot with higher capacity and a boost in spin rate to 5400 rpm. At press time, the company stood by its plans for 5.25-in. designs, despite a stockholder class-action lawsuit claiming that the company overstated the success of the product among OEMs.

3.5-in. drives and interfaces

  Other than Quantum's Bigfoot and a couple of legacy 5.25-in. drives from Seagate nearing the end of life, 3.5-in. drives are now the sole choice for markets ranging from desktop PCs to disk subsystems for supercomputers. By concentrating on the 3.5-in. form factor for high-end drives, manufacturers have maintained balance by increasing rotational speed, linear density, and data rate as heads and read channels emerge to handle the higher speeds. Moreover, manufacturers can integrate the advancements made in high-end 3.5-in. drives into succeeding generations of commodity desktop drives.

  Table 1 oversimplifies the overall market for 3.5-in. drives, providing only three segments. Still, the Table provides a good outline for discussing the drives available. Note that the interface implementation proves to be one of the key distinguishing characteristics among all 3.5-in. drives.

  In the PC market segment, you find a homogenized selection of drives from a number of vendors. You would have to go way beyond the spec sheets to discern any differences in the newest PC drives from different vendors. System designers who simply need the highest quality drives available have to perform extensive application-specific tests to differentiate one product from the next. More likely, designers will make price and availability top priority in choosing a supplier.

  You should make sure that the drive you buy has the proper interface to meet your application requirements. Sometimes, layers of jargon obscure the exact capabilities of the interface. Over the past few years, most people have referred to desktop PC drives as having an IDE (integrated-drive-electronics) or EIDE (enhanced-IDE) interface. Although no such standard truly existed, the disk-drive community continued to enhance and extend the interfaces with different data-transfer modes and other enhanced features. The SFF (Small Form Factor) Committee shepherded many of the enhancements.

  More recently, the SFF Committee has formalized standards for PC disk interfaces and turned them over to the ANSI X3T10 committee. The most accurate and descriptive terminology for such drives today is based on the original name for the drives—ATA (AT Attachment). Basic ATA drives are limited to Mode 1 or 2 PIO (programmed I/O) data transfers at 6 Mbytes/sec. Fast ATA drives support Mode 3 PIO and Mode 1 DMA transfers at 11 Mbytes/sec. ATA-2 drives support Mode 4 PIO and Mode 2 DMA transfers at 16 Mbytes/sec and include support for automatic drive identification. ATA-3 does not increase data rates but tightens the Mode 4 PIO spec to improve the reliability of transfers. ATA-3 also adds simple password security, standardizes power-management features, and adds "SMART" (self-monitoring analysis and report technology), which allows the drive to automatically warn the host about certain types of impending failures.

  Most recently introduced drives for PCs have ATA-2 interfaces. Should you run into a drive with an older interface, it is likely a previous-generation drive. Of course, if the performance, capacity, and price meet your needs, there's nothing wrong with ATA or Fast-ATA drives.

  Drives with ATA-3 interfaces have just begun to appear. For example, Maxtor just announced its Crystalmax family with capacities ranging from 875 Mbytes on one disk to 3.5 Gbytes on four disks. IBM also offers ATA-3 in its Deskstar family of 3.5-in. drives at 2.1 and 3.2 Gbytes, as well as in its newest Travelstar 2.5-in. drives. Despite the availability of ATA-3 drives, few, if any, hosts can currently take advantage of the SMART scheme.

  On the near horizon, yet another enhancement to ATA looms, and it's loaded with more jargon. Quantum developed the technology and dubbed it Ultra DMA/33, and it promises to push ATA interface rates to 33 Mbytes/sec. Ultra DMA/33 doubles the DMA Mode 2 transfer rate by initiating a transfer on both the rising and falling edges of the DMA clock strobe. The new protocol also reduces propagation delay on the interface by letting the target generate the clock strobe when transferring data to the host, whereas previous ATA DMA modes depended on the host to control transfers in both directions. Ultra DMA/33 also includes a CRC to ensure data integrity.

  Quantum has licensed the technology free to competing drive, chip-set, and motherboard vendors. You can expect some product announcements around Comdex, yet host and disk products probably will not ship until spring of 1997. Quantum plans to submit the specification to the ATA standards group, so that, by the time products appear, let's hope for the sake of consistency and clarity that the capabilities inherent in Ultra DMA/43 will fall within the bounds of ATA-4.

SCSI on the desktop

  Although ATA interfaces dominate the desktop market, a few companies ship SCSI products targeted at the desktop. Moreover, OSs such as Windows 95 and NT greatly benefit from the command-queuing capabilities and multidrive support inherent in SCSI (Reference 2).

  Quantum continues to be the only company that offers SCSI on its mainstream desktop drives, and you can buy members of the 3.5-in. Fireball series with either a SCSI or an ATA interface. The SCSI drives typically carry about a $50 price premium. Seagate also offers SCSI on some of its Medalist Pro family members, although even ATA drives in the Medalist Pro family are premium-performance drives relative to the mainstream Medalist family.

  The other market segments for 3.5-in. drives feature much more diversity in drive feature sets, although interface continues to be a key distinguishing factor. Some manufacturers, such as Western Digital, believe that a single drive can serve both the entry-level server and workstation market and the high-end server market. Other companies, such as Quantum, Seagate, and IBM, offer multiple families of SCSI drives targeted at workstations, entry-level servers, and enterprise servers. You should also realize that some of these companies have multiple lines because of acquisitions. For example, Quantum offers the Viking family rooted within Quantum and the Atlas II family originally designed by the Digital Equipment disk operation that Quantum acquired.

  At the mid to high end of the drive market, you probably don't have to worry about drive quality because the vendors are all proven players. You should concentrate on ensuring that the drive you choose meets your application requirements in raw performance, interface performance, and interface choice.

  You can find the best gauge of raw performance in the rotation speed and media transfer-rate specs. SCSI drives for applications other than the desktop should spin at 7200 rpm, and the media-transfer rate should top out at around 140 Mbps. At Comdex, you may see vendors introducing drives that spin at 10,000 rpm and boost the media-transfer rate to 200 Mbps.

  SCSI remains the interface of choice on most mid- to high-end drives, yet all SCSI implementations are far from equal, and SCSI suffers from some of the jargon problems of ATA. A year ago, most drive companies used descriptive terms, such as "SCSI Fast 20," to accurately define the 20-Mbyte/sec performance of the interface. Subsequently, the SCSI Trade Association was formed, and members agreed to use the Ultra SCSI name and derivations to describe products.

  With no guarantee of complete consistency, you can probably expect the following interface when you see the following labels: Fast SCSI-2 should be the baseline drive capable of transferring data at 10 Mbytes/sec across an 8-bit interface. Any drive incapable of this rate is most likely an older generation product. Ultra SCSI or even Ultra SCSI-3 implies a 20-Mbyte/sec rate across an 8-bit interface. Ultra 2 products are not available yet, but vendors could introduce them at Comdex. Ultra 2 implies a 40-Mbyte/sec rate across an 8-bit interface using the new LVDS (low-voltage-differential-signaling) scheme. LVDS is not yet a part of an official ANSI SCSI standard, but vendors have developed a de facto standard LVDS scheme. You will also see the adjective "wide" applied to all the SCSI variations signifying a 16-bit implementation and effective doubling of the transfer rate.

SSA and FC-AL interfaces

  Outside the SCSI realm, you can now buy drives with both the SSA (Serial Storage Architecture) and FC-AL (Fibre Channel Arbitrated Loop) interfaces (Reference 2). SSA drives have been available from IBM for more than a year. Seagate just commenced production shipments of a long-awaited, 9-Gbyte FC-AL Barracuda drive. Seagate also plans to offer FC-AL drives on the 1-in.-high, 4-Gbyte Barracuda 4LP. Some observers assumed that Seagate was not going to offer FC-AL on a 4-Gbyte drive because of the cancellation of a program this year, but Seagate claims it was simply waiting for the 1-in.-high, 4-Gbyte platform. Expect shipments of the 4LP in the first quarter of 1997. Quantum is also shipping evaluation units of an Atlas drive with an FC-AL interface.

  The SSA and FC-AL groups have waged a war of words for more than a year now about which interface will become dominant at the high end. FC-AL offers significant peak data rate advantages over SSA (100 compared to 20 Mbps). SSA, however, inherently includes dual ports that can communicate simultaneously, and its drives in a ring configuration support a concept called "spatial reuse," which allows multiple transfers to occur simultaneously on a ring.

  SSA has won significant market share so far. Dataquest's Devin estimates that $750 million worth of SSA products have been shipped to date. FC-AL advocates, however, claim that IBM captively used most of the products shipped.

  Just days before this issue of EDN went to press, a joint Seagate/IBM press announcement appeared to pave the way for the two groups to merge. The groups announced a concept called FC-EL (Fibre Channel Enhanced Loop) to which both groups would migrate for product shipments commencing in 1998. Presumably, FC-EL would combine the best features of both interfaces into a single standard.

  The announcement actually sent the spin doctors from both sides back to the battlefront. Alex Lieb, vice president of marketing at SSA-adapter vendor Pathlight (Ithaca, NY), claims that FC-EL will essentially place the SSA transport-layer protocol on top of the Fibre Channel physical interface. Lieb claims that transport-layer problems have hampered and delayed FC-AL products, and the SSA interface has proven effective in widespread deployment. Lieb claims that Pathlight and IBM are planning a move from SSA-80 to SSA-160 in the first quarter of next year, boosting peak data rates to 40 Mbytes/sec. They will follow this move with FC-EL products in 1998.

  The FC-AL community, meanwhile, denies any serious transport-layer problems. Watch how quickly Seagate will ramp production of FC-AL drives as an indicator of the validity of any lingering problems. The FCLC (Fibre Channel Loop Community), however, has a different spin on FC-EL. Dal Allan, FCLC facilitator and president of consulting company ENDL, claims that FC-EL is only a concept and that the manufacturers proposing FC-EL have not presented it to the FCLC or ANSI X3T standards groups yet. Allan isn't convinced that FC-EL is more than a name at this point.

  According to Allan, the FCLC and ANSI groups are working on two extensions to FC-AL. FC-AL-2 will allow for single-ring Fibre Channel implementations. Moreover, the new spec will introduce a registered insertion protocol to eliminate arbitration and allow for spatial reuse. The second effort, FC-SL (Fibre Channel Slotted Loop) incorporates the benefits of FC-AL-2 into dual-ring implementations with station-management capabilities and guaranteed isochronous services. According to Allan, a single-drive design can work in an FC-AL-2 or FC-SL environment with the actual differences implemented in the host. He expects products to sample in late 1997. Allan claims that these new FibreChannel schemes already include all of the features that SSA users have found valuable.

  The IBM and Seagate participants behind the FC-EL announcement scheduled a meeting with the FCLC group two weeks after the close of this article. I will include any pertinent updates on the matter in the Leading Edge section of EDN in this or subsequent issues. It appears, however, that the serial-interface battle has not yet ended. Perhaps, activity of the SSA and FCLC trade groups at Comdex will shed more light on the subject.

References

1. Small, Charles H, "Innovative technology boosts disk-drive performance," EDN, April 27, 1995, pg 48.
2. Wright, Maury, "New peripheral interfaces: fast & full of features," EDN, Oct 12, 1995, pg 69.

You can reach Technical Editor Maury Wright at (619) 748-6785, fax (619) 679-1861, ednwright@mcimail.com.

| EDN Access | feedback | subscribe to EDN! |