Drives Get Bigger, Faster, Cheaper

December 1997 / International Bits / Drives Get Bigger, Faster, Cheaper

9.1-GB drives spinning at 7200 rpm will soon enter the mainstream.

Nebojsa Novakovic

Despite many competing technologies, hard disk drives continue to hold their dominant position as the medium of choice for data storage. The price of hard disks continues to plummet, while capacity and performance climb upward. Seagate and Quantum are expected to ship 9-GB Ultra2 SCSI drives in mid-1998. The 3.5-inch-factor drives, which stand just 1 inch high, spin at 10,000 rpm.

The higher the rotational speed of the disk drive motor the greater the data transfer rate, and the lower the seek and latency times when the head switches tracks. Today, most hard disks spin at 5400 rpm, with higher-end models attaining a speed of 7200 rpm. The Seaga te Barr acuda, Quantum Atlas, and Micropolis Tomahawk 7200 drives have become popular in multimedia applications such as digital motion pictures.

Seagate is already shipping its first 10,000-rpm disk drives, the Cheetah family. With formatted data transfer rates peaking at 16.8 MB per second (around 40 percent better than top 7200-rpm hard drives), and prices around 40 percent higher than equivalent 7200-rpm offerings, these 4.5- and 9.1-gigabyte drives are currently limited to niche market applications. Early next year, however, Seagate will sell a new drive, the Medalist Pro 9140 , that has a 9.1-GB capacity and spins at 7200 rpm. Street price will be about $495, according to Denise Lippert, product marketing manager.

Not all agree that high-end 10,000-rpm drives are a viable business right now. David Rawcliffe, marketing director of Quantum Asia-Pacific, says Quantum will offer 10,000-rpm models in mid-1998, when 10,000-rpm drives might have 15 percent of the high-end hard disk market. Quantum's first 10,000-rpm entries are expected to hold 9 GB and 18 GB, with either Ultra2 SCSI or Fibre Channel interfaces. Rawcliffe also expects 10,000 rpm to remain the top hard disk rotational speed until early next decade.

Magnetoresistive (MR) heads, used by IBM and Quantum, helped achieve the next level of capacity by separating the read and write functions into two physically distinct heads. An inductive head, optimized for writing information, is integrated with the MR structure optimized for reading. MR heads are usually coupled with partial response maximum likelihood (PRML) channel recording for resolving data pulses from the disk surface. While the initial troubles in manufacturing MR heads limited their use to very high-end drives, the yields have improved in the past two years. Major vendors like IBM, Quantum, and Seagate use them in many mainstream IDE drives.

The next step, Giant MR (GMR), enabling recording density in tens of gigabit s per square inch, is expected to arrive next year in some 3.5-inch drives with capacities well above 10 GB. Western Digital is already shipping UltraATA IDE devices with 2.16 GB formatted capacity per platter (6.4 GB for a three-platter hard disk) using MR head technology. In 1998, MR heads are expected to overtake the old inductive ones in terms of number of units shipped, analysts say.

Besides GMR and improvements to PRML, like IBM's PRML with digital filtering, new technologies like multilevel decision feedback equalization (MDFE) are expected to help achieve even higher capacities. Developed by Singapore-based Data Storage Institute, MDFE is expected to surpass PRML and its derivatives in recording density and performance.

On the high end, dual-head parallel drives, pioneered by Seagate, didn't make it in the market. Seagate dropped its 2-GB Barracuda 2HP two years ago, with no plans to use the technology in the future. While dual-head parallel operation doubles data transfer speed for both r eads and writes, similar speedups can be achieved by striping two identical drives on the same or different SCSI connections. Data can be interleaved on a word or sector basis, depending on whether your target application is graphics- or database-oriented. Windows NT directly supports disk striping, joining the drives into a single volume.

Makers of high-end drives do not need to follow the small incremental steps in capacity seen in IDE drives. Instead of 2.1 GB, 2.5 GB, 3.2 GB, 4.3 GB, etc., the high-end arena has the simple capacity-doubling rule : 2.2 GB, 4.5 GB, and 9 GB. The 18-GB Quantum Atlas III is presently the highest-capacity 3.5-inch drive.

UltraSCSI, in its 20-MBps 8-bit and 40-MBps 16-bit versions, is now the dominant high-end mass storage interface. Many high-end drives are available in the most popular versions of this interface, including combinations of narrow, wide, single connector attachment (SCA), and differential UltraSCSI. Early next year, volume shipm ents of the first Ultra2 SCSI drives with new LVD-Link (low voltage differential) transmitter technology will start. The 16-bit version of Ultra2 SCSI not only doubles the transfer rate to 80 MBps, it also increases the maximum cable length from 1.5 to 12 meters. The first PCI controller to provide Ultra2 Wide SCSI with LVD-Link is Symbios's 53C895 Ultra2 Wide PCI/SCSI I/O processor. The company is expected to offer a dual-channel Ultra2 SCSI processor early next year on a 64-bit PCI bus, the follow-on to its 53C876 dual-channel Ultra Wide SCSI PCI processor.

For the next year, there are more design improvements scheduled by the ANSI XT310 committee, including compact packaging, tripled connector density, and "smart silicon." With Ultra4 Wide SCSI, bus bandwidth will increase to 160 MBps.

The Fibre Channel-Arbitrated Loop (FC-AL) provides serial interface performance of 100 MBps per loop, with up to two loops connected to each storage device. This unprecedented performance rate can be sustained on links up to 10 kilometers long, with as many as 126 devices in a single chain. Adaptec and Symbios were among the first to offer FC-AL PCI host adapters earlier this year.

Quantum's Rawcliffe says the 1998 high-end focus will still be on Ultra2 SCSI with LVD, as the performance difference between it and FC-AL is not so drastic, and there will still be around a 20 percent price differential between otherwise identical drives in Ultra2 SCSI and FC-AL versions. Also, FC-AL controllers will still be substantially more expensive, even with single-chip implementations like the Adaptec AIC 1160 and the Symbios FC920. Adaptec product manager Adam Zagorski states that the company expects "two distinct markets to develop for Ultra2 SCSI and Fibre Channel. People will use Ultra2 to support their legacy SCSI drives."

A year ago, Seagate and IBM agreed to merge Seagate-backed FC-AL and IBM-backed Serial Storage Architecture (SSA) into one serial storage interface standard dubbed Fibre Channel-Enhanced Loop ( FC-EL), combining SSA reliability and FC-AL performance. It could become the dominant long-term high-end mass storage interface, with performance moving toward 800 MBps and beyond. Like FC-AL, FC-EL is both a network and a storage interface standard, supporting Internet, ATM, and SCSI protocols, among others. The first FC-EL products are expected in 1999.

But don't expect high-end drives that predict and warn you of their failure. Self-Monitoring, Analysis, and Reporting Technology (SMART) is not planned for most workstation and server systems next year.

Desktop Hard Disk Evolution

illustration_link (21 Kbytes)

PC hard disk capacity is expected to climb to 36 GB by the year 2000.

Gettin' Better All the Time

photo_link (74 Kbytes)

Seagate's new 9.1-GB drive is expected to sell for around $495 when introduced early next year.