Table of Contents For This Issue
|How Computers Work, Part I|
August 2001 Vol.5 Issue 3 |
Page(s) 18-23 in print issue
Hard Drives Made Easy
Investigating These Mundane Marvels
NOTE: This article has been updated from the print version.|
A hard disk drive, or hard drive, is instrumental to your PC’s performance. Without its reliable, quick storage capacity, you would probably have to start your computer using data from a diskette or CD-ROM. Operating systems such as Windows 2000 and applications such as your Internet browser would take much longer to load. Even worse, retrieving your e-mail could feel like waiting for real mail, as Internet servers would have to rely on slower media. Non-volatile memory would be a relatively fast alternative, but is much more expensive.
A hard drive magnetically reads and writes information to and from spinning platters (hard disks) coated with magnetic recording media via tiny read/write heads. An actuator arm moves the heads among the platters, making any piece of stored data accessible in milliseconds. The five main steps in a hard drive’s operation are input, processing, writing, reading, and output.
Input. The computer’s OS (operating system) keeps track of a hard drive’s data with a directory system, such as the FAT32 (32-bit file allocation table), and tells the drive where to record new information. The OS sends new data from the motherboard through a controller (built-in or on an adapter card) and a ribbon data cable to the hard drive.
The computer employs an interface (a standardized connection between the motherboard and a device), such as SCSI (Small Computer System Interface) or EIDE (Enhanced Integrated Drive Electronics, or its variants Ultra ATA [Ultra Advanced Technology Attachment] or Ultra DMA [Ultra Direct Memory Access]), to communicate with the hard drive. A drive’s interface can almost always handle a faster data stream than the drive can supply. However, a new drive could get bogged down by an older, slower interface. Several fast drives could similarly overwhelm a recent interface if they shared a bus or older RAID (redundant array of independent disks).
The data reaches the hard drive’s cache buffer, which is memory that temporarily stores the data in segmented areas until it can be processed. Modern cache buffers range from 2MB to 4MB, and up to 16MB for full-motion video applications.
Meanwhile, the drive’s platters spin rapidly at between 5,400 and 15,000rpm (revolutions per minute) on a spindle inside the drive’s case. This spindle, or shaft, keeps the platters a precise distance from each other and is rigid enough to avoid errors even after physical shocks of up to 400G (400 times the Earth’s gravity) in certain desktop drives. Some spindles ride on fluid, rather than ball, bearings for reduced noise and longer life. The HDB (hydrodynamic bearings) option for Maxtor’s (formerly Quantum’s) Fireball Plus AS is an example. The controller adjusts the speed of the spindle motor and platters.
Hard drives are generally low-level formatted by the manufacturer. This involves mapping each platter into tracks (concentric rings) and sectors (usually 512-byte sections of a track). Formatting writes system data to the beginning of each sector for purposes of timing, head positioning (see below), and identification. The user high-level formats the drive, writing file storage structures to the sectors before use.
A cylinder is a grouping of the same track on each platter. This concept is more useful than tracks because the actuator arm moves all the heads at once. To read a track on one platter, all the heads will be in that cylinder. The computer’s OS keeps track of data by cylinder, head, and sector, and it organizes related sectors into clusters of 2KB to 32KB each.
Processing. The PCB (printed circuit board) mounted under the hard drive’s chassis holds the drive’s control electronics and cache buffer. The PCB’s DSP (digital signal processor) converts the serial digital data from the computer into analog data that can be stored on the platters. A circuit called the read channel encodes the write data so it can be stored more efficiently and, as its name implies, will be used later to decode read data.
Writing. The write heads create the electromagnetic signals to be stored on the platters’ magnetic coating. The write heads are paired closely alongside the read heads, which sometimes share writing duties by acting as one of the two poles the write heads require to function.
Write heads resemble tiny “C” shapes made of magnetic material wound with fine wire. They are inductive, meaning that a current passed through them causes a magnetic field. Head manufacturers use thin-film technology (the depositing of extremely thin layers of metal) and other fabrication techniques. Generally, the smaller the gap between the two poles of the write head, the higher the signal frequency it can produce and the denser the data that can be written to the platter (called areal density).
Each head literally flies over the platter’s polished surface at a height as little as 1/5000th of the thickness of a human hair in certain drives. Air currents above the spinning disk lift the head’s slider (aerodynamic mounting) like the wing of an airplane. Modern hard drives often employ the SMART (Self-Monitoring, Analysis, and Reporting Technology) system to predict failures and warn the user before they occur.
Here are the primary steps involved in writing data to disk:
1. The controller instructs the actuator arm to move the heads over to the correct track, or cylinder, of the platters.
2. The actuator arm pivots when precise amounts of voltage are sent to its coiled magnets (or voice-coil servo). The arm’s range of motion between the inner and outer edges of the platters, combined with the platters’ spins, allows the heads to access any area of the hard disk in milliseconds.
3. The system information in each sector’s header helps the drive position its heads. The computer’s OS ensures the controller writes the data to unused, preferably contiguous sectors.
Each platter has a read/write head for each usable side. A preamplifier ensures adequate signal strength to and from the heads. The platters, actuator arm, and their related motors and servos are all sealed in the hard drive’s case by a gasketed cover to prevent contamination. A finely filtered vent hole keeps the atmospheric pressure inside the drive equal to that of the surrounding air. There may be other passive filters in a hard drive that remove microscopic particles from the airflow of the platters.
Reading. When the PC’s OS requests data from a certain location on the hard drive, the controller tells the actuator arm to swing to the proper cylinder. The read head positioned above the correct track of that cylinder will read the data and send it to the cache buffer.
Modern read heads are almost always the GMR (giant magnetoresistive) type, sometimes called spin-valve heads. Earlier MR (magnetoresistive) heads consisted of a tiny chunk of nickel-iron alloy. In contrast, GMR heads sandwich a conductive but non-magnetic layer of copper between two small blocks of nickel-iron. GMR heads are much more sensitive than MR heads, have a better signal-to-noise ratio, and can read denser data. The magnetoresistive property of both types of heads means they function by having a constant sense current (or bias current) passed through them. When the heads pass over a change in a magnetic field, such as the data on a hard disk, their inherent resistance changes, and so does the amount of sense current they let pass. The hard drive interprets the stored data by the patterns in the sense currents.
Output. The read channel deciphers the sense current’s data, using the cache buffer as temporary storage. The DSP converts it back to digital information. Current read channels use EPRML (Extended Partial Response Maximum Likelihood) or MEEPRML (Modified Enhanced EPRML). These improvements upon PRML, which digitally samples the analog data read from a hard drive and predicts its peaks and valleys, accommodate higher areal densities.
Most error-correction routines, such as CRC (cyclic redundancy check), are run at this time, although some also monitor the writing process. Finally, the requested data is sent to the PC through the data cable.
The average seek time , or the average time it takes the actuator arm to position the heads over a random track, advertised with hard drives doesn’t convey much real-world information. A hypothetical 5,400rpm drive could have the same seek time as a 15,000rpm drive, but with all else being equal, there’s no way the 5,400rpm model could access data as quickly. If you add the average seek time to the drive’s rotational latency, you’ll get an average access time, telling you how long it takes the drive to find a particular piece of data. For example, a 7,200rpm drive takes 4.17ms to spin its platters halfway around. Add 4.17ms to the average seek time—say, 8.5ms—and you’ll get an average access time of 12.67ms.
A hard drive’s interface speed doesn’t tell you much about how fast the drive is, either. The interface speed, such as 100MBps for Ultra ATA/100, merely tells you the maximum amount of data per second the drive’s cache buffer can send to the computer. However, the real measure of a drive’s sequential speed is how fast it can actually read or write data to or from the platters. These average data transfer rates (also called throughput) incorporate the maximum read/write speeds near the outer edges of the platters (hard drives write the outer tracks first) and the slower read/write speeds toward the inner tracks.
A drive’s average access time is most important for random access applications such as word processing, spreadsheets, and servers with multiple users; its average data transfer rates are more important for sequential data access such as audio and video files.
When the hard drive shuts down, its disks stop spinning and the heads will land on the platters or other mechanisms. Modern drives automatically pull the actuator arm to a land zone (a dedicated, unused cylinder) or a ramp (such as IBM’s Load/Unload System) away from the platters. This ensures the heads do not accidentally gouge any sectors containing user data. Land zones are often textured to keep the heads from sticking to the platters.
Hard Drive History. Today’s high-capacity desktop hard drives, at a minimum cost of 1.1 cents per megabyte and falling, represent the best balance of speed and cost-effectiveness in a random-access storage device. However, the technology goes back 45 years.
1956. IBM was the first company to ship a hard drive. The RAMAC 305 (Random Access Method of Accounting and Control 305) could store 5MB on 50 24-inch disks, and it cost about $50,000 ($10,000 per megabyte). It was as large as two refrigerators.
1973. IBM introduced the first Winchester hard drive, the 3340, in which the heads flew above its two 30MB (30-30) platters. It was nicknamed after IBM’s laboratory in Winchester, England, and/or the popular .30-30 lever-action rifle, according to different sources. Modern hard drives with flying heads are still called Winchester drives.
1980. The first thin-film inductive heads enabled the IBM 3380 to read or write an unprecedented 3MBps (megabytes per second).
1984. Western Digital developed the first hard drive controller board for the IBM PC/AT.
1985. Western Digital collaborated with Compaq and Control Data to develop the IDE 40-pin interface.
1990. IBM developed the PRML (Partial Response Maximum Likelihood) read channel.
1991. IBM and Fujitsu introduced drives with MR heads. These read heads could function regardless of platter speed and detected weaker (and denser) signals than inductive read heads.
The average hard drive’s capacity reached 145MB, according to IBM. The average cost was $5.32 per megabyte.
1992. Seagate announced the first 7,200rpm hard drive, the Barracuda.
1993. Seagate set an industry record by shipping its 50 millionth hard drive.
1994. Western Digital pioneered the EIDE interface specification, which allowed faster data transfer rates of up to 11MBps and the use of drives larger than 528MB.
Quantum started using Compaq’s SMART in hard drives.
1996. IBM achieved an areal density of 1Gb (gigabit; a billion bits) per square inch. Quantum and Intel co-invented the Ultra ATA/33 (16-bit, 33MBps data transfer rate) interface.
Fujitsu introduced actuator arms with integrated head wiring. This improved the aerodynamics of the assembly, allowing lower head flying heights.
Seagate’s Cheetah, the first 10,000rpm drive, debuted in October. One month later, Seagate became the first manufacturer to ship its 100 millionth hard drive.
1997. In December, IBM introduced the first hard drive with GMR heads: a 16.8GB drive with 4.1Gb/square inch (512.5MB/square inch) areal density.
1998. Quantum debuted the first Ultra2 SCSI (40MBps, 8-bit; 80MBps, 16-bit) interface and co-invented Ultra ATA/66 (66MBps, 16-bit) with Intel.
1999. In June, IBM unveiled the world’s smallest hard drive, the 1-inch 340MB Microdrive. The tiny drive fits in the same slot as CF+ (Compact Flash+) Type II memory modules.
Over time, the form factors, or sizes of the physical assembly, of hard drives decreased significantly. Common form factors were:
•14-inch and 8-inch in the 1960s and 1970s
•5.25-inch in the 1980s and 1990s
•3.5-inch from the 1980s until today
•3-inch and 2.5-inch for mobile computers from the late 1980s until today
•1.8-inch, from Toshiba, for mobile devices
•1.3-inch, a short-lived size in the early 1990s
•1-inch (IBM’s Microdrive) in the late 1990s.
As hard drive capacities increased, OSes and BIOSes (Basic Input/Output Systems) sometimes imposed limits on the size of hard drives they would support. Designers overcame barriers at 128MB, 528MB, 2GB, and 8.4GB by changing the way files were stored, transferred, and identified, as well as how the computer’s BIOS interacted with the drive itself.
Today. Today’s desktop drives pull a wider variety of duties than ever before. E-mail, Web content, video games, digital photos, video, and audio files fuel demand for bigger and faster storage.
Hard drives are moving beyond computers. Drives of various form factors appear in video-editing suites, laser printers, photocopiers, and high-end digital cameras. Another hard-drive powered gadget gaining popularity is the DVR (digital video recorder). DVRs, such as those from TiVo (http://www.tivo.com), are essentially tapeless VCRs that can continue recording your TV even as you rewind or play back earlier footage. A DVR’s hard drive is quick enough to read video from one area as it continues writing incoming data to another.
The largest desktop and mobile drive capacities today are 181.55GB and 48GB, respectively. Desktop units available in new PCs vary between 20GB and 80GB, while typical notebook drives fall between 10GB and 30GB. Today, 3.5-inch drives are universal in desktops and servers, and 2.5-inch drives are common in notebooks.
Today’s hard drive interfaces shipping in new desktop PCs are Ultra ATA/100 (an extension of EIDE) at a theoretical maximum throughput of 100MBps. Ultra2 SCSI and Ultra160 SCSI at 160MBps appear in servers and performance PCs, but Seagate already supports the new Ultra320 SCSI interface (320MBps) in its Cheetah X15 36LP and other drives. Manufacturers also use the FC-AL (Fibre Channel-Arbitrated Loop; up to 2Gbps) interface for server drives and other applications.
In 1999, John Best, vice president of technology in IBM’s Storage Systems Division, said that areal density is the fastest growing performance frontier in hard drive technology. It’s still true today. “It’s really the thing that drives the ability to build huge capacities, as well as to drive [the] cost-per-bit down,” he said.
Manufacturers, such as IBM and Fujitsu, improve areal density with smaller, higher-frequency heads that fly closer to smoother platters, and better magnetic material in the platters themselves. “All new designs shrink the size of the individual bit while providing a viable signal-to-noise ratio and improved product life,” said Mike Chenery, vice president of architecture development for Fujitsu.
The problem is that the current recording media on hard disk platters limits areal density to roughly 40Gb per square inch or less. The write heads need to write bits of data in such tiny areas, the media can’t reliably maintain the bits’ magnetic orientation. The media essentially “forgets” some of its data. This is called the superparamagnetic effect.
In May 2001, however, IBM announced a new type of media it says could yield up to 100Gb per square inch areal densities by 2003. The company’s new AFC (antiferromagnetically-coupled) media uses a layer of the elemental metal ruthenium, just three atoms thick, between two layers of magnetic recording media. The ruthenium layer keeps the two recording layers polarized opposite each other, simultaneously keeping the entire three-layer structure very thin. The net effect is that AFC media can “remember” tinier bits of data, thereby allowing higher areal densities.
Future. IBM’s Microdrive ($459 for 1GB) is the most likely hard drive to be used in portable devices, such as digital music players and consumer digital cameras, as its price comes down. It’s simply more economical than flash memory cards of the same capacity. “The disk drive still provides the optimum choice for any digital storage that requires greater than 1 GB of capacity,” said Chenery. “As more consumer equipment moves to digital, disk drives become increasingly essential components.”
Handheld computers, such as the HP Jornada 720 (with a PC Card adapter), and gadgets, such as Iomega’s FotoShow photo viewer, can also use the Microdrive. Other candidates for miniature hard drives include GPS (global positioning satellite) devices, smartphones, and onboard PCs in cars.
On a larger scale, Perception Digital chose Seagate’s U Series 5 hard drive for its PDHercules MP3 jukebox appliance. Meanwhile, Western Digital and Seagate are supplying hard drives for Microsoft’s Xbox game console debuting this fall.
Performance. IBM and most observers agree areal density is the quickest key to performance, but there are other hurdles besides the superparamagnetic effect.
“As you go toward higher areal densities,” Scranton said, “every one of the critical dimensions has to be reduced.” This includes the heads’ flying height above the platter. Scranton said the new IBM Travelstar 30GN drives’ heads fly less than 20 nanometers (one billionth of a meter) above the platter—between 1/2500th and 1/5000th of the thickness of a human hair.
Another critical dimension for areal density is the gap between the elements of the read/write heads. Manufacturers are also examining more sensitive head designs, such as CPP (current-perpendicular-to-plane) heads. CPP heads are similar to MR and GMR heads, except that the sense current flows through the layers of material in the head (perpendicular) rather than across them (parallel). Whereas GMR heads have a conductive copper sheet between their layers, MTJ (magnetic tunnel junction, or TMR [tunnel MR]) heads have an insulating layer thin enough to actually let some current through in the presence of certain magnetic fields.
“Basically, the electron current jumps, or tunnels, across the thin insulator, and its ability to jump or tunnel depends upon the relative direction of magnetizations of the (head’s) two electrodes,” Scranton says. “We believe . . . that it could make an even more sensitive transducer than GMR heads.”
In 1999, Best predicted radically changing the recording media along with refined scaling methods for other components could yield an entire terabit (1 trillion bits) in a square inch. Projecting from the consensual 100% annual growth rate in areal density, hard drives could reach one tera-bit/square inch by 2007, and terabyte capacity (1,000GB) in 2003.
As an example, Seagate’s colossal 181.55GB Barracuda 180 has twelve 15.6Gb/square inch platters. If the Barracuda 180 had 100Gb/square inch disks, it could top 1.15TB (terabytes), or 1,150GB. Optically-assisted Winchester (OAW) drives, in which optical sensors help position the heads, could unlock even higher areal densities.
Meanwhile, Serial ATA will likely succeed EIDE. This new 150MBps interface offers thinner data cables (for better airflow inside the computer) and easier setup.
Hard drives are breaking into new applications as quickly as they are breaking old performance records. In the Information Age, everybody needs data storage. Hard drives are simply the best way to achieve it.
by Marty Sems
View the graphics that accompany this article.
(NOTE: These pages are PDF (Portable Document Format) files. You will need Adobe Acrobat Reader to view these pages. Download Adobe Acrobat Reader)
View additional graphics that accompany this article.
(NOTE: These pages are PDF (Portable Document Format) files. You will need Adobe Acrobat Reader to view these pages. Download Adobe Acrobat Reader)