Internals of a 2.5-inch SATA hard disk drive
|24 December 1954; 62 years ago (1954-12-24)|
|IBM team led by Rey Johnson|
A hard disk drive (HDD), hard disk, hard drive or fixed disk is a data storage device that uses magnetic storage to store and retrieve digital information using one or more rigid rapidly rotating disks (platters) coated with magnetic material. The platters are paired with magnetic heads, usually arranged on a moving actuator arm, which read and write data to the platter surfaces. Data is accessed in a random-access manner, meaning that individual blocks of data can be stored or retrieved in any order and not only sequentially. HDDs are a type of non-volatile storage, retaining stored data even when powered off.
Introduced by IBM in 1956, HDDs became the dominant secondary storage device for general-purpose computers by the early 1960s. Continuously improved, HDDs have maintained this position into the modern era of servers and personal computers. More than 200 companies have produced HDDs historically, though after extensive industry consolidation most current units are manufactured by Seagate, Toshiba, and Western Digital. HDD unit shipments and sales revenues are declining, though production (exabytes per year) is growing. Flash memory has a growing share of the market for secondary storage, in the form of solid-state drives (SSDs). SSDs have higher data-transfer rates, higher areal storage density, better reliability, and much lower latency and access times. Though SSDs have higher cost per bit, they are replacing HDDs where speed, power consumption, small size, and durability are important.
The primary characteristics of an HDD are its capacity and performance. Capacity is specified in unit prefixes corresponding to powers of 1000: a 1-terabyte (TB) drive has a capacity of 1,000 gigabytes (GB; where 1 gigabyte = 1 billion bytes). Typically, some of an HDD's capacity is unavailable to the user because it is used by the file system and the computer operating system, and possibly inbuilt redundancy for error correction and recovery. Performance is specified by the time required to move the heads to a track or cylinder (average access time) plus the time it takes for the desired sector to move under the head (average latency, which is a function of the physical rotational speed in revolutions per minute), and finally the speed at which the data is transmitted (data rate).
The two most common form factors for modern HDDs are 3.5-inch, for desktop computers, and 2.5-inch, primarily for laptops. HDDs are connected to systems by standard interface cables such as PATA (Parallel ATA), SATA (Serial ATA), USB or SAS (Serial attached SCSI) cables.
|3.75 megabytes||12 terabytes||3.2-million-to-one|
|Physical volume||68 cubic feet (1.9 m)||2.1 cubic inches (34 cm)||56,000-to-one|
|Weight||2,000 pounds (910 kg)||2.2 ounces (62 g)||15,000-to-one|
|Average access time||approx. 600 milliseconds||2.5 ms to 10 ms; RW RAM dependent||about
|Price||US$9,200 per megabyte (1961)||US$0.032 per gigabyte by 2015||300-million-to-one|
|Data density||2,000 bits per square inch||1.3 terabits per square inch in 2015||650-million-to-one|
|Average lifespan||~2000 hrs MTBF||~2500000 hrs MTBF||1250-to-one|
Hard disk drives were introduced in 1956 as data storage for an IBM real-time transaction processing computer, and were developed for use with general-purpose mainframe and minicomputers. The first IBM drive, the 350 RAMAC in 1956, was approximately the size of two medium-sized refrigerators and stored five million six-bit characters (3.75 megabytes) on a stack of 50 disks.
In 1962 the IBM 350 RAMAC disk storage unit was superseded by the IBM 1301 disk storage unit, which consisted of 50 platters, each about 1/8-inch thick and 24 inches in diameter. Whereas the IBM 350 used only two read/write heads which were pneumatically actuated and moved in two dimensions, the 1301 was one of the first disk storage units to use an array of heads, one per platter, moving as a single unit. Cylinder-mode read/write operations were supported, and the heads flew about 250 micro-inches (about 6 µm) above the platter surface. Motion of the head array depended upon a binary adder system of hydraulic actuators which assured repeatable positioning. The 1301 cabinet was about the size of three home refrigerators placed side by side, storing the equivalent of about 21 million eight-bit bytes. Access time was about a quarter of a second.
Also in 1962, IBM introduced the model 1311 disk drive, which was about the size of a washing machine and stored two million characters on a removable disk pack. Users could buy additional packs and interchange them as needed, much like reels of magnetic tape. Later models of removable pack drives, from IBM and others, became the norm in most computer installations and reached capacities of 300 megabytes by the early 1980s. Non-removable HDDs were called "fixed disk" drives.
Some high-performance HDDs were manufactured with one head per track (e.g. IBM 2305 in 1970) so that no time was lost physically moving the heads to a track. Known as fixed-head or head-per-track disk drives they were very expensive and are no longer in production.
In 1973, IBM introduced a new type of HDD code-named "Winchester". Its primary distinguishing feature was that the disk heads were not withdrawn completely from the stack of disk platters when the drive was powered down. Instead, the heads were allowed to "land" on a special area of the disk surface upon spin-down, "taking off" again when the disk was later powered on. This greatly reduced the cost of the head actuator mechanism, but precluded removing just the disks from the drive as was done with the disk packs of the day. Instead, the first models of "Winchester technology" drives featured a removable disk module, which included both the disk pack and the head assembly, leaving the actuator motor in the drive upon removal. Later "Winchester" drives abandoned the removable media concept and returned to non-removable platters.
Like the first removable pack drive, the first "Winchester" drives used platters 14 inches (360 mm) in diameter. A few years later, designers were exploring the possibility that physically smaller platters might offer advantages. Drives with non-removable eight-inch platters appeared, and then drives that used a 5 ⁄ in (130 mm) form factor (a mounting width equivalent to that used by contemporary floppy disk drives). The latter were primarily intended for the then-fledgling personal computer (PC) market.
As the 1980s began, HDDs were a rare and very expensive additional feature in PCs, but by the late 1980s their cost had been reduced to the point where they were standard on all but the cheapest computers.
Most HDDs in the early 1980s were sold to PC end users as an external, add-on subsystem. The subsystem was not sold under the drive manufacturer's name but under the subsystem manufacturer's name such as Corvus Systems and Tallgrass Technologies, or under the PC system manufacturer's name such as the Apple ProFile. The IBM PC/XT in 1983 included an internal 10 MB HDD, and soon thereafter internal HDDs proliferated on personal computers.
External HDDs remained popular for much longer on the Apple Macintosh. Many Macintosh computers made between 1986 and 1998 featured a SCSI port on the back, making external expansion simple. Older compact Macintosh computers did not have user-accessible hard drive bays (indeed, the Macintosh 128K, Macintosh 512K, and Macintosh Plus did not feature a hard drive bay at all), so on those models external SCSI disks were the only reasonable option for expanding upon any internal storage.
The 2011 Thailand floods damaged the manufacturing plants and impacted hard disk drive cost adversely between 2011 and 2013.
Driven by ever increasing areal density since their invention, HDDs have continuously improved their characteristics; a few highlights are listed in the table above. At the same time, market application expanded from mainframe computers of the late 1950s to most mass storage applications including computers and consumer applications such as storage of entertainment content.
A modern HDD records data by magnetizing a thin film of ferromagnetic material on a disk. Sequential changes in the direction of magnetization represent binary data bits. The data is read from the disk by detecting the transitions in magnetization. User data is encoded using an encoding scheme, such as run-length limited encoding, which determines how the data is represented by the magnetic transitions.
A typical HDD design consists of a spindle that holds flat circular disks, also called platters, which hold the recorded data. The platters are made from a non-magnetic material, usually aluminum alloy, glass, or ceramic, and are coated with a shallow layer of magnetic material typically 10–20 nm in depth, with an outer layer of carbon for protection. For reference, a standard piece of copy paper is 0.07–0.18 millimeters (70,000–180,000 nm).
The platters in contemporary HDDs are spun at speeds varying from 4,200 rpm in energy-efficient portable devices, to 15,000 rpm for high-performance servers. The first HDDs spun at 1,200 rpm and, for many years, 3,600 rpm was the norm. As of December 2013, the platters in most consumer-grade HDDs spin at either 5,400 rpm or 7,200 rpm.
Information is written to and read from a platter as it rotates past devices called read-and-write heads that are positioned to operate very close to the magnetic surface, with their flying height often in the range of tens of nanometers. The read-and-write head is used to detect and modify the magnetization of the material passing immediately under it.
In modern drives, there is one head for each magnetic platter surface on the spindle, mounted on a common arm. An actuator arm (or access arm) moves the heads on an arc (roughly radially) across the platters as they spin, allowing each head to access almost the entire surface of the platter as it spins. The arm is moved using a voice coil actuator or in some older designs a stepper motor. Early hard disk drives wrote data at some constant bits per second, resulting in all tracks having the same amount of data per track but modern drives (since the 1990s) use zone bit recording – increasing the write speed from inner to outer zone and thereby storing more data per track in the outer zones.
In modern drives, the small size of the magnetic regions creates the danger that their magnetic state might be lost because of thermal effects, thermally induced magnetic instability which is commonly known as the "superparamagnetic limit". To counter this, the platters are coated with two parallel magnetic layers, separated by a 3-atom layer of the non-magnetic element ruthenium, and the two layers are magnetized in opposite orientation, thus reinforcing each other. Another technology used to overcome thermal effects to allow greater recording densities is perpendicular recording, first shipped in 2005, and as of 2007 the technology was used in many HDDs.
In 2004, a new concept was introduced to allow further increase of the data density in magnetic recording, using recording media consisting of coupled soft and hard magnetic layers. That so-called exchange spring media, also known as exchange coupled composite media, allows good writability due to the write-assist nature of the soft layer. However, the thermal stability is determined only by the hardest layer and not influenced by the soft layer.
A typical HDD has two electric motors; a spindle motor that spins the disks and an actuator (motor) that positions the read/write head assembly across the spinning disks. The disk motor has an external rotor attached to the disks; the stator windings are fixed in place. Opposite the actuator at the end of the head support arm is the read-write head; thin printed-circuit cables connect the read-write heads to amplifier electronics mounted at the pivot of the actuator. The head support arm is very light, but also stiff; in modern drives, acceleration at the head reaches 550 g.
The actuator is a permanent magnet and moving coil motor that swings the heads to the desired position. A metal plate supports a squat neodymium-iron-boron (NIB) high-flux magnet. Beneath this plate is the moving coil, often referred to as the voice coil by analogy to the coil in loudspeakers, which is attached to the actuator hub, and beneath that is a second NIB magnet, mounted on the bottom plate of the motor (some drives have only one magnet).
The voice coil itself is shaped rather like an arrowhead, and made of doubly coated copper magnet wire. The inner layer is insulation, and the outer is thermoplastic, which bonds the coil together after it is wound on a form, making it self-supporting. The portions of the coil along the two sides of the arrowhead (which point to the actuator bearing center) then interact with the magnetic field of the fixed magnet. Current flowing radially outward along one side of the arrowhead and radially inward on the other produces the tangential force. If the magnetic field were uniform, each side would generate opposing forces that would cancel each other out. Therefore, the surface of the magnet is half north pole and half south pole, with the radial dividing line in the middle, causing the two sides of the coil to see opposite magnetic fields and produce forces that add instead of canceling. Currents along the top and bottom of the coil produce radial forces that do not rotate the head.
The HDD's electronics control the movement of the actuator and the rotation of the disk, and perform reads and writes on demand from the disk controller. Feedback of the drive electronics is accomplished by means of special segments of the disk dedicated to servo feedback. These are either complete concentric circles (in the case of dedicated servo technology), or segments interspersed with real data (in the case of embedded servo technology). The servo feedback optimizes the signal to noise ratio of the GMR sensors by adjusting the voice-coil of the actuated arm. The spinning of the disk also uses a servo motor. Modern disk firmware is capable of scheduling reads and writes efficiently on the platter surfaces and remapping sectors of the media which have failed.
Error rates and handling
Modern drives make extensive use of error correction codes (ECCs), particularly Reed–Solomon error correction. These techniques store extra bits, determined by mathematical formulas, for each block of data; the extra bits allow many errors to be corrected invisibly. The extra bits themselves take up space on the HDD, but allow higher recording densities to be employed without causing uncorrectable errors, resulting in much larger storage capacity. For example, a typical 1 TB hard disk with 512-byte sectors provides additional capacity of about 93 GB for the ECC data.
In the newest drives, as of 2009, low-density parity-check codes (LDPC) were supplanting Reed–Solomon; LDPC codes enable performance close to the Shannon Limit and thus provide the highest storage density available.
Typical hard disk drives attempt to "remap" the data in a physical sector that is failing to a spare physical sector provided by the drive's "spare sector pool" (also called "reserve pool"), while relying on the ECC to recover stored data while the amount of errors in a bad sector is still low enough. The S.M.A.R.T (Self-Monitoring, Analysis and Reporting Technology) feature counts the total number of errors in the entire HDD fixed by ECC (although not on all hard drives as the related S.M.A.R.T attributes "Hardware ECC Recovered" and "Soft ECC Correction" are not consistently supported), and the total number of performed sector remappings, as the occurrence of many such errors may predict an HDD failure.
The "No-ID Format", developed by IBM in the mid-1990s, contains information about which sectors are bad and where remapped sectors have been located.
Only a tiny fraction of the detected errors ends up as not correctable. For example, specification for an enterprise SAS disk (a model from 2013) estimates this fraction to be one uncorrected error in every 10 bits, and another SAS enterprise disk from 2013 specifies similar error rates. Another modern (as of 2013) enterprise SATA disk specifies an error rate of less than 10 non-recoverable read errors in every 10 bits. An enterprise disk with a Fibre Channel interface, which uses 520 byte sectors to support the Data Integrity Field standard to combat data corruption, specifies similar error rates in 2005.
The worst type of errors are those that go unnoticed, and are not even detected by the disk firmware or the host operating system. These errors are known as silent data corruption, some of which may be caused by hard disk drive malfunctions.
The rate of areal density advancement was similar to Moore's law (doubling every two years) through 2010: 60% per year during 1988–1996, 100% during 1996–2003 and 30% during 2003–2010. Gordon Moore (1997) called the increase "flabbergasting," while observing later that growth cannot continue forever. Price improvement decelerated to −12% per year during 2010–2017, as the growth of areal density slowed. The rate of advancement for areal density slowed to 10% per year during 2010–2016, and there was difficulty in migrating from perpendicular recording to newer technologies.
Areal density is the inverse of bit cell size, so an increase in areal density corresponds to a decrease in bit cell size. In 2013, a production desktop 3 TB HDD (with four platters) would have had an areal density of about 500 Gbit/in which would have amounted to a bit cell comprising about 18 magnetic grains (11 by 1.6 grains). Since the mid-2000s areal density progress has increasingly been challenged by a superparamagnetic trilemma involving grain size, grain magnetic strength and ability of the head to write. In order to maintain acceptable signal to noise smaller grains are required; smaller grains may self-reverse (electrothermal instability) unless their magnetic strength is increased, but known write head materials are unable to generate a magnetic field sufficient to write the medium. Several new magnetic storage technologies are being developed to overcome or at least abate this trilemma and thereby maintain the competitiveness of HDDs with respect to products such as flash memory-based solid-state drives (SSDs).
In 2013, Seagate introduced one such technology, shingled magnetic recording (SMR). Additionally, SMR comes with design complexities that may cause reduced write performance. Other new recording technologies that, as of 2016, still remain under development include heat-assisted magnetic recording (HAMR), microwave-assisted magnetic recording (MAMR), two-dimensional magnetic recording (TDMR), bit-patterned recording (BPR), and "current perpendicular to plane" giant magnetoresistance (CPP/GMR) heads.
The rate of areal density growth has dropped below the historical Moore's law rate of 40% per year, and the deceleration is expected to persist through at least 2020. Depending upon assumptions on feasibility and timing of these technologies, the median forecast by industry observers and analysts for 2020 and beyond for areal density growth is 20% per year with a range of 10–30%. The achievable limit for the HAMR technology in combination with BPR and SMR may be 10 Tbit/in, which would be 20 times higher than the 500 Gbit/in represented by 2013 production desktop HDDs. As of 2015, HAMR HDDs have been delayed several years, and are expected in 2018. They require a different architecture, with redesigned media and read/write heads, new lasers, and new near-field optical transducers.
The capacity of a hard disk drive, as reported by an operating system to the end user, is smaller than the amount stated by the manufacturer, which has several reasons: the operating system using some space, use of some space for data redundancy, and space use for file system structures. The difference in capacity reported in true SI-based units vs. binary prefixes can lead to a false impression of missing capacity.
Modern hard disk drives appear to their host controller as a contiguous set of logical blocks, and the gross drive capacity is calculated by multiplying the number of blocks by the block size. This information is available from the manufacturer's product specification, and from the drive itself through use of operating system functions that invoke low-level drive commands.
The gross capacity of older HDDs is calculated as the product of the number of cylinders per recording zone, the number of bytes per sector (most commonly 512), and the count of zones of the drive. Some modern SATA drives also report cylinder-head-sector (CHS) capacities, but these are not physical parameters because the reported values are constrained by historic operating system interfaces. The C/H/S scheme has been replaced by logical block addressing (LBA), a simple linear addressing scheme that locates blocks by an integer index, which starts at LBA 0 for the first block and increments thereafter. When using the C/H/S method to describe modern large drives, the number of heads is often set to 64, although a typical hard disk drive, as of 2013, has between one and four platters.
In modern HDDs, spare capacity for defect management is not included in the published capacity; however, in many early HDDs a certain number of sectors were reserved as spares, thereby reducing the capacity available to the operating system.
For RAID subsystems, data integrity and fault-tolerance requirements also reduce the realized capacity. For example, a RAID 1 array has about half the total capacity as a result of data mirroring, while a RAID 5 array with x drives loses 1/x of capacity (which equals to the capacity of a single drive) due to storing parity information. RAID subsystems are multiple drives that appear to be one drive or more drives to the user, but provide fault tolerance. Most RAID vendors use checksums to improve data integrity at the block level. Some vendors design systems using HDDs with sectors of 520 bytes to contain 512 bytes of user data and eight checksum bytes, or by using separate 512-byte sectors for the checksum data.
Some systems may use hidden partitions for system recovery, reducing the capacity available to the end user.
The presentation of a hard disk drive to its host is determined by the disk controller. The actual presentation may differ substantially from the drive's native interface, particularly in mainframes or servers. Modern HDDs, such as SAS and SATA drives, appear at their interfaces as a contiguous set of logical blocks that are typically 512 bytes long, though the industry is in the process of changing to the 4,096-byte logical blocks layout, known as the Advanced Format (AF).
The process of initializing these logical blocks on the physical disk platters is called low-level formatting, which is usually performed at the factory and is not normally changed in the field. As a next step in preparing an HDD for use, high-level formatting writes partition and file system structures into selected logical blocks to make the remaining logical blocks available to the host's operating system and its applications. The file system uses some of the disk space to structure the HDD and organize files, recording their file names and the sequence of disk areas that represent the file. Examples of partition mapping scheme include Master boot record (MBR) and GUID Partition Table (GPT). Examples of data structures stored on disk to retrieve files include the File Allocation Table (FAT) in the DOS file system and inodes in many UNIX file systems, as well as other operating system data structures (also known as metadata). As a consequence, not all the space on an HDD is available for user files, but this system overhead is usually negligible.
|100 GB||100,000,000,000||107,374,182,400||7.37%||93.1 GB, 95,367 MB||100 GB|
|1 TB||1,000,000,000,000||1,099,511,627,776||9.95%||931 GB, 953,674 MB||1,000 GB, 1,000,000 MB|
The total capacity of HDDs is given by manufacturers in SI-based units such as gigabytes (1 GB = 1,000,000,000 bytes) and terabytes (1 TB = 1,000,000,000,000 bytes). The practice of using SI-based prefixes (denoting powers of 1,000) in the hard disk drive and computer industries dates back to the early days of computing; by the 1970s, "million", "mega" and "M" were consistently used in the decimal sense for drive capacity. However, capacities of memory (RAM, ROM) and CDs are traditionally quoted using a binary interpretation of the prefixes, i.e. using powers of 1024 instead of 1000.
Internally, computers do not represent either hard disk drive or memory capacity in powers of 1,024, but reporting it in this manner is a convention. The Microsoft Windows family of operating systems uses the binary convention when reporting storage capacity, so an HDD offered by its manufacturer as a 1 TB drive is reported by these operating systems as a 931 GB HDD. Mac OS X 10.6 ("Snow Leopard") uses decimal convention when reporting HDD capacity. The default behavior of the df command-line utility on Linux is to report the HDD capacity as a number of 1024-byte units.
The difference between the decimal and binary prefix interpretation caused some consumer confusion and led to class action suits against HDD manufacturers. The plaintiffs argued that the use of decimal prefixes effectively misled consumers while the defendants denied any wrongdoing or liability, asserting that their marketing and advertising complied in all respects with the law and that no class member sustained any damages or injuries.
HDD price per byte improved at the rate of −40% per year during 1988–1996, −51% per year during 1996–2003, and −34% per year during 2003–2010. The price improvement decelerated to −13% per year during 2011–2014, as areal density increase slowed and the 2011 Thailand floods damaged manufacturing facilities.
|3.5||Current||146||101.6||19, 25.4 or 26.1||12 TB (December 2016)||5-8||1,149|
|2.5||Current||100||69.85||5, 7, 9.5, 12.5, 15 or 19||5 TB (2016)||5||1,000|
|1.8||Obsolete||78.5||54||5 or 8||320 GB (2009)||2||220|
|5.25 (FH)||Obsolete||203||146||82.6||47 GB (1998)||14||3.36|
|5.25 (HH)||Obsolete||203||146||41.4||19.3 GB (1998)||4||4.83|
|1.3||Obsolete||?||43||?||40 GB (2007)||1||40|
|1 (CFII/ZIF/IDE-Flex)||Obsolete||?||42||?||20 GB (2006)||1||20|
|0.85||Obsolete||32||24||5||8 GB (2004)||1||8|
IBM's first hard drive, the IBM 350, used a stack of fifty 24-inch platters and was of a size comparable to two large refrigerators. In 1962, IBM introduced its model 1311 disk, which used six 14-inch (nominal size) platters in a removable pack and was roughly the size of a washing machine. This became a standard platter size and drive form-factor for many years, used also by other manufacturers. The IBM 2314 used platters of the same size in an eleven-high pack and introduced the "drive in a drawer" layout, although the "drawer" was not the complete drive.
Later drives were designed to fit entirely into a chassis that would mount in a 19-inch rack. Digital's RK05 and RL01 were early examples using single 14-inch platters in removable packs, the entire drive fitting in a 10.5-inch-high rack space (six rack units). In the mid-to-late 1980s the similarly sized Fujitsu Eagle, which used (coincidentally) 10.5-inch platters, was a popular product.
Such large platters were never used with microprocessor-based systems. With increasing sales of microcomputers having built in floppy-disk drives (FDDs), HDDs that would fit to the FDD mountings became desirable. Thus HDD Form factors, initially followed those of 8-inch, 5.25-inch, and 3.5-inch floppy disk drives. Because there were no smaller floppy disk drives, smaller HDD form factors developed from product offerings or industry standards.
As of 2012, 2.5-inch and 3.5-inch hard disks were the most popular sizes.
By 2009, all manufacturers had discontinued the development of new products for the 1.3-inch, 1-inch and 0.85-inch form factors due to falling prices of flash memory, which has no moving parts.
While these sizes are customarily described by an approximately correct figure in inches, actual sizes have long been specified in millimeters.
Time to access data
The factors that limit the time to access the data on an HDD are mostly related to the mechanical nature of the rotating disks and moving heads. Seek time is a measure of how long it takes the head assembly to travel to the track of the disk that contains data. Rotational latency is incurred because the desired disk sector may not be directly under the head when data transfer is requested. These two delays are on the order of milliseconds each. The bit rate or data transfer rate (once the head is in the right position) creates delay which is a function of the number of blocks transferred; typically relatively small, but can be quite long with the transfer of large contiguous files. Delay may also occur if the drive disks are stopped to save energy.
An HDD's Average Access Time is its average seek time which technically is the time to do all possible seeks divided by the number of all possible seeks, but in practice is determined by statistical methods or simply approximated as the time of a seek over one-third of the number of tracks.
Defragmentation is a procedure used to minimize delay in retrieving data by moving related items to physically proximate areas on the disk. Some computer operating systems perform defragmentation automatically. Although automatic defragmentation is intended to reduce access delays, performance will be temporarily reduced while the procedure is in progress.
Time to access data can be improved by increasing rotational speed (thus reducing latency) or by reducing the time spent seeking. Increasing areal density increases throughput by increasing data rate and by increasing the amount of data under a set of heads, thereby potentially reducing seek activity for a given amount of data. The time to access data has not kept up with throughput increases, which themselves have not kept up with growth in bit density and storage capacity.
Average seek time ranges from under 4 ms for high-end server drives to 15 ms for mobile drives, with the most common mobile drives at about 12 ms and the most common desktop type typically being around 9 ms. The first HDD had an average seek time of about 600 ms; by the middle of 1970s, HDDs were available with seek times of about 25 ms. Some early PC drives used a stepper motor to move the heads, and as a result had seek times as slow as 80–120 ms, but this was quickly improved by voice coil type actuation in the 1980s, reducing seek times to around 20 ms. Seek time has continued to improve slowly over time.
Some desktop and laptop computer systems allow the user to make a tradeoff between seek performance and drive noise. Faster seek rates typically require more energy usage to quickly move the heads across the platter, causing louder noises from the pivot bearing and greater device vibrations as the heads are rapidly accelerated during the start of the seek motion and decelerated at the end of the seek motion. Quiet operation reduces movement speed and acceleration rates, but at a cost of reduced seek performance.
Latency is the delay for the rotation of the disk to bring the required disk sector under the read-write mechanism. It depends on rotational speed of a disk, measured in revolutions per minute (rpm). Average rotational latency is shown in the table on the right, based on the statistical relation that the average latency in milliseconds for such a drive is one-half the rotational period. Average latency (in milliseconds) is computed as 30,000 divided by rotational speed (in rpm).
Data transfer rate
As of 2010, a typical 7,200-rpm desktop HDD has a sustained "disk-to-buffer" data transfer rate up to 1,030 Mbit/s. This rate depends on the track location; the rate is higher for data on the outer tracks (where there are more data sectors per rotation) and lower toward the inner tracks (where there are fewer data sectors per rotation); and is generally somewhat higher for 10,000-rpm drives. A current widely used standard for the "buffer-to-computer" interface is 3.0 Gbit/s SATA, which can send about 300 megabyte/s (10-bit encoding) from the buffer to the computer, and thus is still comfortably ahead of today's disk-to-buffer transfer rates. Data transfer rate (read/write) can be measured by writing a large file to disk using special file generator tools, then reading back the file. Transfer rate can be influenced by file system fragmentation and the layout of the files.
HDD data transfer rate depends upon the rotational speed of the platters and the data recording density. Because heat and vibration limit rotational speed, advancing density becomes the main method to improve sequential transfer rates. Higher speeds require a more powerful spindle motor, which creates more heat. While areal density advances by increasing both the number of tracks across the disk and the number of sectors per track, only the latter increases the data transfer rate for a given rpm. Since data transfer rate performance tracks only one of the two components of areal density, its performance improves at a lower rate.
Other performance considerations include quality-adjusted price, power consumption, audible noise, and both operating and non-operating shock resistance.
The Federal Reserve Board has a quality-adjusted price index for large-scale enterprise storage systems including three or more enterprise HDDs and associated controllers, racks and cables. Prices for these large-scale storage systems improved at the rate of ‒30% per year during 2004–2009 and ‒22% per year during 2009–2014.
Access and interfaces
HDDs are accessed over one of a number of bus types, including as of 2011 parallel ATA (PATA, also called IDE or EIDE; described before the introduction of SATA as ATA), Serial ATA (SATA), SCSI, Serial Attached SCSI (SAS), and Fibre Channel. Bridge circuitry is sometimes used to connect HDDs to buses with which they cannot communicate natively, such as IEEE 1394, USB and SCSI.
Modern HDDs present a consistent interface to the rest of the computer, no matter what data encoding scheme is used internally. Typically a DSP in the electronics inside the HDD takes the raw analog voltages from the read head and uses PRML and Reed–Solomon error correction to decode the sector boundaries and sector data, then sends that data out the standard interface. That DSP also watches the error rate detected by error detection and correction, and performs bad sector remapping, data collection for Self-Monitoring, Analysis, and Reporting Technology, and other internal tasks.
Modern interfaces connect an HDD to a host bus interface adapter (today typically integrated into the "south bridge") with one data/control cable. Each drive also has an additional power cable, usually direct to the power supply unit.
- Small Computer System Interface (SCSI), originally named SASI for Shugart Associates System Interface, was standard on servers, workstations, Commodore Amiga, Atari ST and Apple Macintosh computers through the mid-1990s, by which time most models had been transitioned to IDE (and later, SATA) family disks. The range limitations of the data cable allows for external SCSI devices.
- Integrated Drive Electronics (IDE), later standardized under the name AT Attachment (ATA, with the alias P-ATA or PATA (Parallel ATA) retroactively added upon introduction of SATA) moved the HDD controller from the interface card to the disk drive. This helped to standardize the host/controller interface, reduce the programming complexity in the host device driver, and reduced system cost and complexity. The 40-pin IDE/ATA connection transfers 16 bits of data at a time on the data cable. The data cable was originally 40-conductor, but later higher speed requirements for data transfer to and from the HDD led to an "ultra DMA" mode, known as UDMA. Progressively swifter versions of this standard ultimately added the requirement for an 80-conductor variant of the same cable, where half of the conductors provides grounding necessary for enhanced high-speed signal quality by reducing cross talk.
- EIDE was an unofficial update (by Western Digital) to the original IDE standard, with the key improvement being the use of direct memory access (DMA) to transfer data between the disk and the computer without the involvement of the CPU, an improvement later adopted by the official ATA standards. By directly transferring data between memory and disk, DMA eliminates the need for the CPU to copy byte per byte, therefore allowing it to process other tasks while the data transfer occurs.
- Fibre Channel (FC) is a successor to parallel SCSI interface on enterprise market. It is a serial protocol. In disk drives usually the Fibre Channel Arbitrated Loop (FC-AL) connection topology is used. FC has much broader usage than mere disk interfaces, and it is the cornerstone of storage area networks (SANs). Recently other protocols for this field, like iSCSI and ATA over Ethernet have been developed as well. Confusingly, drives usually use copper twisted-pair cables for Fibre Channel, not fibre optics. The latter are traditionally reserved for larger devices, such as servers or disk array controllers.
- Serial Attached SCSI (SAS). The SAS is a new generation serial communication protocol for devices designed to allow for much higher speed data transfers and is compatible with SATA. SAS uses a mechanically identical data and power connector to standard 3.5-inch SATA1/SATA2 HDDs, and many server-oriented SAS RAID controllers are also capable of addressing SATA HDDs. SAS uses serial communication instead of the parallel method found in traditional SCSI devices but still uses SCSI commands.
- Serial ATA (SATA). The SATA data cable has one data pair for differential transmission of data to the device, and one pair for differential receiving from the device, just like EIA-422. That requires that data be transmitted serially. A similar differential signaling system is used in RS485, LocalTalk, USB, FireWire, and differential SCSI.
Integrity and failure
Due to the extremely close spacing between the heads and the disk surface, HDDs are vulnerable to being damaged by a head crash – a failure of the disk in which the head scrapes across the platter surface, often grinding away the thin magnetic film and causing data loss. Head crashes can be caused by electronic failure, a sudden power failure, physical shock, contamination of the drive's internal enclosure, wear and tear, corrosion, or poorly manufactured platters and heads.
The HDD's spindle system relies on air density inside the disk enclosure to support the heads at their proper flying height while the disk rotates. HDDs require a certain range of air densities in order to operate properly. The connection to the external environment and density occurs through a small hole in the enclosure (about 0.5 mm in breadth), usually with a filter on the inside (the breather filter). If the air density is too low, then there is not enough lift for the flying head, so the head gets too close to the disk, and there is a risk of head crashes and data loss. Specially manufactured sealed and pressurized disks are needed for reliable high-altitude operation, above about 3,000 m (9,800 ft). Modern disks include temperature sensors and adjust their operation to the operating environment. Breather holes can be seen on all disk drives – they usually have a sticker next to them, warning the user not to cover the holes. The air inside the operating drive is constantly moving too, being swept in motion by friction with the spinning platters. This air passes through an internal recirculation (or "recirc") filter to remove any leftover contaminants from manufacture, any particles or chemicals that may have somehow entered the enclosure, and any particles or outgassing generated internally in normal operation. Very high humidity present for extended periods of time can corrode the heads and platters.
For giant magnetoresistive (GMR) heads in particular, a minor head crash from contamination (that does not remove the magnetic surface of the disk) still results in the head temporarily overheating, due to friction with the disk surface, and can render the data unreadable for a short period until the head temperature stabilizes (so called "thermal asperity", a problem which can partially be dealt with by proper electronic filtering of the read signal).
When the logic board of a hard disk fails, the drive can often be restored to functioning order and the data recovered by replacing the circuit board with one of an identical hard disk. In the case of read-write head faults, they can be replaced using specialized tools in a dust-free environment. If the disk platters are undamaged, they can be transferred into an identical enclosure and the data can be copied or cloned onto a new drive. In the event of disk-platter failures, disassembly and imaging of the disk platters may be required. For logical damage to file systems, a variety of tools, including fsck on UNIX-like systems and CHKDSK on Windows, can be used for data recovery. Recovery from logical damage can require file carving.
A common expectation is that hard disk drives designed and marketed for server use will fail less frequently than consumer-grade drives usually used in desktop computers. However, two independent studies by Carnegie Mellon University and Google found that the "grade" of a drive does not relate to the drive's failure rate.
A 2011 summary of research, into SSD and magnetic disk failure patterns by Tom's Hardware summarized research findings as follows:
- Mean time between failures (MTBF) does not indicate reliability; the annualized failure rate is higher and usually more relevant.
- Magnetic disks do not have a specific tendency to fail during early use, and temperature has only a minor effect; instead, failure rates steadily increase with age.
- S.M.A.R.T. warns of mechanical issues but not other issues affecting reliability, and is therefore not a reliable indicator of condition.
- Failure rates of drives sold as "enterprise" and "consumer" are "very much similar", although these drive types are customized for their different operating environments.
- In drive arrays, one drive's failure significantly increases the short-term risk of a second drive failing.
Manufacturers and sales
More than 200 companies have manufactured HDDs over time. But consolidations have concentrated production into just three manufacturers today: Western Digital, Seagate, and Toshiba.
Worldwide revenue for disk storage declined 4% per year, from a peak of $38 billion in 2012 to $27 billion in 2016. Production of HDDs grew 16% per year, from 335 exabytes in 2011 to 693 exabytes in 2016. Shipments declined 7% per year during this time period, from 620 million units to 425 million. Seagate and Western Digital each have 40–45% of unit shipments, while Toshiba has 13–17%. The average sales price for the two largest manufacturers was $60 per unit in 2015.
Competition from SSD
The maximum areal storage density for flash memory used in SSDs is 2.8 Tbit/in in laboratory demonstrations as of 2016, and the maximum for HDDs is 1.5 Tbit/in. The areal density of flash memory is doubling every two years, similar to Moore's law (40% per year) and faster than the 10–20% per year for HDDs. As of 2016, maximum capacity was 10 terabytes for an HDD, and 15 terabytes for an SSD. HDDs were used in 70% of the desktop and notebook computers produced in 2016, and SSDs were used in 30%. The usage share of HDDs is declining and could drop below 50% in 2018–2019 according to one forecast, because SSDs are replacing smaller-capacity (less than one-terabyte) HDDs in desktop and notebook computers and MP3 players.
The market for silicon-based flash memory (NAND) chips, used in SSDs and other applications, is growing rapidly. Worldwide revenue grew 12% per year during 2011–2016. It rose from $22 billion in 2011 to $39 billion in 2016, while production grew 46% per year from 19 exabytes to 120 exabytes.
External hard disk drives
External hard disk drives typically connect via USB; variants using USB 2.0 interface generally have slower data transfer rates when compared to internally mounted hard drives connected through SATA. Plug and play drive functionality offers system compatibility and features large storage options and portable design. As of March 2015, available capacities for external hard disk drives ranged from 500 GB to 10 TB.
External hard disk drives are usually available as pre-assembled integrated products, but may be also assembled by combining an external enclosure (with USB or other interface) with a separately purchased drive. They are available in 2.5-inch and 3.5-inch sizes; 2.5-inch variants are typically called portable external drives, while 3.5-inch variants are referred to as desktop external drives. "Portable" drives are packaged in smaller and lighter enclosures than the "desktop" drives; additionally, "portable" drives use power provided by the USB connection, while "desktop" drives require external power bricks.
Features such as biometric security or multiple interfaces (for example, Firewire) are available at a higher cost. There are pre-assembled external hard disk drives that, when taken out from their enclosures, cannot be used internally in a laptop or desktop computer due to embedded USB interface on their printed circuit boards, and lack of SATA (or Parallel ATA) interfaces.
Hard disk drives are traditionally symbolized as a stylized stack of platters or as a cylinder, and are as such found in various diagrams; sometimes, they are depicted with small lights to indicate data access. In most modern graphical user environments (GUIs), hard disk drives are represented by an illustration or photograph of the drive enclosure.
- Hard Disk Drives Encyclopedia
- Video showing an opened HD working
- Average seek time of a computer disk
- Timeline: 50 Years of Hard Drives
- HDD from inside: Tracks and Zones. How hard it can be?
- Hard disk hacking – firmware modifications, in eight parts, going as far as booting a Linux kernel on an ordinary HDD controller board
- Hiding Data in Hard Drive’s Service Areas, February 14, 2013, by Ariel Berkman
- Rotary Acceleration Feed Forward (RAFF) Information Sheet, Western Digital, January 2013
- PowerChoice Technology for Hard Disk Drive Power Savings and Flexibility, Seagate Technology, March 2010
- Shingled Magnetic Recording (SMR), HGST, Inc., 2015
- The Road to Helium, HGST, Inc., 2015
- Research paper about perspective usage of magnetic photoconductors in magneto-optical data storage.