Market Overview ? Hard Disks: From Stone Age to Storage
Hard disks have come a long way since the dawn of computer time. The smaller they get, the more they can hold. Roger Gann takes a look at thelatest generation
If there?s one maxim that holds true, it?s ?data expands to fill the space available?. You can never be too thin, too rich or have too much hard disk storage space. It seems that no matter how much hard disk space you have, people will find lots of creative ways to fill it. This unfortunate truism leads to an eternal search for more and better ways of maximising storage options. Which is just as well, as the hard disk after-market for those wishing to expand the storage capacity of their PCs remains pretty healthy.
In the old days, a hard disk was something that was not dissimilar in size to a fridge. In the 70s it was the cause of some excitement when the latest 200Mb fridges were delivered to a mainframe site. Let us not forget that the hard disk wasn?t introduced to the PC until two years after its launch. And unless you could afford an arm and a leg you couldn?t even think of buying one. That?s all changed.
Price trends in the disk storage market are downward a little like walking off the edge of the white cliffs of Dover is downward. Capacities will double again later this year and the cost per Mb of storage will continue to plummet. According to the US newsletter Disk/Trend, disk storage prices were originally about $2.50 per Mb. Today, it?s about 10 cents per Mb with the price perhaps dropping to two cents per Mb by the end of the century.
These price reductions are really due to the massive improvements in hard disk technology over the past five years. Research into materials and the mechanical operation of drives has thrown up faster disks with better specifications.
One of the first places to look for improvement is in head assembly. The newest head technology is called magneto- resistive (MR), which is designed to support media with very high recording densities in the range of one to two billion bits per square inch (Bpsi) compared with the densities of less than one tenth of this achievable with current head technologies.
Unlike current head technologies, all of which are basically tiny inductive electro- magnets, MR technology uses a different approach for reading, based on a special material whose electrical resistance changes in the presence of a magnetic field. A small stripe of MR material is deposited on the head structure and, as it passes over the magnetic patterns on the disk it senses the strength of the magnetic field and creates electrical pulses corresponding to flux reversals.
This mechanism cannot be used for writing, so a conventional thin-film inductive write element is deposited alongside the MR stripe. MR head technology began to appear in drive designs in 1994, with wider acceptance and inclusion in more designs by 1995. It?s largely due to the use of the MR heads, coupled with PRML read channels (see next page) that the 1Gb drive has become a reality.
Once you have played with the head, the disk platter needs looking at. Rotation speeds have been increasing as manufacturing tolerances have improved. The very first hard disks used to rotate at 3,600rpm. Over the years, this has in-creased, first to 5,400rpm, later to 7,200rpm. Now drives that rotate at 10,000rpm, such as the Cheetah range from Seagate, are starting to appear.
These are the first 3.5in drives to attain this spin speed. Intended for network and multimedia server-level PCs, the Cheetah drives will offer average seek times as low as 7.5ms (milliseconds) and average sustained data-transfer rates of up to 14Mbps. The Cheetah?s four-platter design holds 4.6Gb of data in a 1in-high form factor, while the eight-platter, 1.6in-high version holds 9.1Gb. By rotating at 10,000rpm, the average latency time (amount of time necessary on average to rotate a platter to the point where the head is positioned at the beginning of the data stream to be read) is reduced to 2.99ms, nearly 40 per cent better than 7,200rpm drives.
However, it isn?t a simple job. The transition from 7,200 to 10,000rpm will be a real challenge to hard disk manufacturers. Hard disk spindle motors typically draw four, five or six Watts ? at 10,000rpm, you?re looking at 10 to 14 Watts. The motors will run hotter, and the bearing lubricants will want to run out of the bearings at those speeds and temperatures.
The faster and hotter they are, the more need there is for some sort of system status monitoring and that?s available as well. Introduced in 1996, the Smart (self-monitoring analysis and reporting technology) system uses internal hard disk drive monitoring technology, along with a graphical user interface to alert users if an impending device failure is detected. The Norton Smart Doctor application, developed jointly by Quantum and Symantec, provides a user interface for desktop PCs that enables users to realise the Smart system benefits of increased data protection and enhanced system reliability. The Smart system is an industry standard for both Scsi and ATA interfaces and is currently awaiting approval as a standard for DMI (desktop management interface).
Clearly, software isn?t psychic and Smart can?t predict things like power failures. However, mechanical failures, which are mainly predictable failures, account for 60 per cent of all drive failures. With Smart technology, more predictable failures will be predicted, and data loss will be avoided. Most larger EIDE drives now feature Smart protection.
While the software isn?t psychic, there are technologies that try to be and one of them sits on the data read/write system on the hard disk. PRML (partial response maximum likelihood) was first used in digital communications and was adopted by IBM for disk drives as long ago as 1990. A hard disk?s read channel performs the data encoding and conversions needed to write data to the drive and then read back the data with a high degree of accuracy.
Traditionally, drives have used analogue peak detection read channels. However, digital filtering read channels, such as PRML, are starting to supersede analogue peak detection read channels because they allow the drive to read and write data packed closer together on the plat- ter surface.
During a write operation, an analogue peak detection read channel converts binary, digital data into an analogue signal, which the drive?s read/write head uses to cause magnetic flux changes on the platter surface. During a read operation, the read/write head detects the magnetic flux changes from the data and generates an analogue read-back signal, in the form of a wave, that it transmits to the read channel. The read channel analyses the incoming signal to determine the high/ positive peaks and the low/ negative peaks (the peaks in the analogue wave occurring where the signal is strongest). Finally, it decodes each of these peaks into a single bit of digital data.
The problem with this technique is that as data densities increase, the analogue signal peaks start to merge. To offset this problem, hard drives employ a data encoding scheme during write ope- rations that separates analogue signal peaks.
The downside is that this encoding effectively puts a ceiling on the data density and hence the storage capacity of a drive. Using sophisticated digital coding and filtering techniques, PRML read channels sample the analogue signal wave at a number of points, as opposed to just at the high and low peaks. It then analyses the samples to determine the shape of the read-back signal, and thus can interpret the high and low peaks that represent data bits very accurately. Although using the same read/write heads and media, the use of PRML technology can deliver a 25 per cent improvement in data bit density while achieving the same, low bit error rate as analogue peak detection. Another advantage of PRML is that since the bit density is higher, the drive?s internal data transfer rate will be higher.
With all the technologies producing more efficient hardware, the receiving end of the chain needs some improvement. A new drive protocol designed to double the data transfer rate of PC-based hard drives has been co-developed by Quantum and Intel. The new protocol, called Ultra DMA/33, will double the burst data transfer rate of ATA/IDE hard drives to 33Mbps. Currently, both the DMA 2 and the PIO Mode 4 support burst-data transfer rates of 16.6Mbps. Top- of-the-line disk drives with low latencies, fast seek times, and rotational speeds of 7,400 rpm, offer sustained data transfer rates between 7.5 and 8.9Mbps.
In addition to speed improvements, the protocol brings new data integrity capabilities to the ATA/IDE interface. Improved timing margins and the use of cyclical redundancy check (CRC), a data protection verification not implemented in legacy ATA modes, help to ensure the integrity of transferred data.
The new protocol will be backward compatible with the existing Fast ATA/Enhanced IDE specification and will work with existing devices, including ATAPI CD-Rom drives. Because of this feature, installed PCs without the Ultra DMA/33 capability can use new disk drives in legacy ATA modes at transfer rates up to 16.6Mbps. To take advantage of the high-speed 33Mbps protocol, PC users in the installed base can purchase an Ultra DMA/33 PCI adaptor card.
Ultra DMA will add very little, if anything, to the cost of drives, the additional cost of new ASICs being compensated by the reduction in the size of the cache buffer now required.
A number of alternatives to the existing EIDE/Scsi interfaces are waiting in the wings, all offering vastly better throughputs. First is fibre channel, which offers data transfer rates starting at 100bps, 25m cable lengths and up to 126 devices.
Next is serial storage architecture or SSA. This isn?t so fast, delivering data transfer rates starting at 80Mbps, 20m cable lengths and up to 25 devices supported. Bringing up the rear is P1394 (or Firewire). This has through-puts of about 12.5Mbps and features a low-cost and ergonomically designed connector.
Storage density researchers at IBM recently demonstrated components that read and write data at a density of 5Gbsi. That?s nearly three times the density of the highest-density disk drive, IBM?s Travelstar VP, a 2.5in format drive for portables, which clocks up a data density of 1.44Gbsi. To put it another way, at 5Gbsi density, this disk drive could hold more than 6Gb of data.
New media increases in areal density are particularly difficult to achieve ? the higher the density, the fainter the signal and the closer the heads must fly to the media surface. One way to reduce the fly height of the heads is to create a smoother platter, free of flaws that could cause a disk crash. Glass appears to hold the most promise, and has seen limited commercial use already. However, glass has drawbacks ? it?s relatively fragile and it?s pricey.
Another alternative is to let the head ride directly on the surface of the platter. But this contact unavoidably incurs friction. Such ?contact? recording technology might use a liquid lubricant, or ?wet disk? technology, or low-friction materials that wouldn?t wear out. But, wet disk technology is some way off.
On the market
Seagate, having absorbed rival Conner Peripherals, is now the hard disk maker with the largest market share, (27.5 per cent), followed by Quantum with 22 per cent and Western Digital with 18.4 per cent. Seagate, however, dominates the high-end (4Gb to 9Gb) sector with a 59 per cent market share. Dataquest further predicts: Growth rates dropping down from just under 30 per cent in 1995 to a forecast 21.3 per cent this year. In unit volume terms that means an increase from 90 million units in 1995, through 106 million in 1996 to 132.6 million in 1997.
The strongest growth will be in the 2Gb-plus sector. By the year?s end the most popular drive will be 2.5Gb, shipping to OEMs for the same price as last year?s 1.2Gb drives.
In the high-end market, for drives used in servers, workstations and data warehousing, 9Gb drives will reach good volumes, but 4Gb will be the hot spot.
The fastest growing desktop technology will be Ultra DMA, delivering 33Mbps throughput with higher reliability.
On spindle speeds, the high-end market, now dominated by 7,200rpm units, will begin moving to 10,000rpm drives, like Seagate?s new Chee- tah, in 1998, and should be evenly balanced between the two speeds by 1999. Scsi is also set to take a greater share of the hard disk market: as more corporates make the move to Windows NT4, they are likely to make greater use of Scsi rather than EIDE drives. This is due to the fact that Scsi is able to handle multiple simultaneous I/O requests and EIDE cannot do the same.