Magnetic Disk

Part of Data Storage

Platters, read/write heads, and random access — how spinning magnetic disks became the backbone of computing storage for sixty years.

Why This Matters

The magnetic hard disk drive transformed computing from a batch-processing technology into an interactive one. Where tape required sequential access and drums had modest capacity, the hard disk combined large capacity with fast random access. From IBM’s RAMAC of 1956 through the solid-state transition of the 2010s, the hard disk was the defining storage technology of the computing era.

Understanding disk operation matters even after the semiconductor storage transition. Billions of hard disk drives exist in salvageable form worldwide. They work on simple physical principles. A functioning hard disk drive requires no exotic materials to operate — only the original manufactured components, which are extremely durable when not subjected to physical shock or head crashes.

Additionally, the organizational concepts pioneered by disk storage — tracks, sectors, cylinders, file allocation tables, directory structures — underlie all subsequent storage technologies including flash and optical. Understanding disks means understanding the entire storage ecosystem.

Physical Architecture

A hard disk drive contains one or more circular platters coated with magnetic oxide or thin-film metallic magnetic material. Platters are typically aluminium (early drives) or glass (modern drives for reduced vibration), precisely balanced, and spun at 5,400 to 15,000 RPM by a brushless DC motor at the center hub.

Recording surfaces: Each platter has two recording surfaces (top and bottom). Each surface has its own read/write head. A drive with 4 platters has 8 surfaces and 8 heads. All heads are mounted on a single actuator assembly and move in unison — you cannot position one head independently of the others.

Tracks: Each surface is divided into thousands of concentric circular tracks. Track 0 is at the outer edge; the innermost tracks have the highest track number. Modern drives have tens of thousands of tracks per surface; early drives had a few hundred. Track density (tracks per inch of radius) is one of the major parameters determining capacity.

Sectors: Each track is divided into fixed-length sectors. The traditional sector size is 512 bytes; drives from around 2010 onward use 4,096 byte (4 KB) sectors. Each sector also contains header information (identifying the sector address) and error correction bytes appended to the data payload.

Cylinders: All tracks at the same radial position across all surfaces form a cylinder. Since all heads move together, all tracks in a cylinder can be accessed without moving the actuator. This matters for performance: data that logically belongs together should be stored in the same cylinder to minimize head movement.

Zoned bit recording (ZBR): Outer tracks are physically longer than inner tracks. Early disks wasted this by recording the same number of sectors on every track. ZBR stores more sectors on outer tracks than inner tracks, significantly increasing total capacity. Most drives since the early 1990s use ZBR.

The Air Bearing: How Heads Fly

The read/write head does not touch the disk surface. It flies on a microscopic air cushion — typically 5–10 nanometers above the surface in modern drives, a few micrometers in older drives. This air bearing is created by the wedge-shaped slider that carries the head: as the disk spins beneath it, air is dragged along and compressed under the leading edge of the slider, generating lift that exactly balances the spring force pushing the head toward the surface.

The advantage of the air bearing: since the head never touches, there is no friction wear and the head-disk interface can last for billions of revolutions. The disadvantage: if the air bearing collapses — because the drive is operating in very thin air (high altitude), because the disk surface has a dust particle, or because the disk is spun up too slowly — the head contacts the disk surface in what is called a head crash, destroying both the head and the disk surface beneath it.

This is why hard drives are sealed: a single dust particle between head and disk would cause an immediate crash. The internal atmosphere is filtered to sub-micron cleanliness. When a hermetically sealed drive is opened outside a cleanroom, contamination enters and the drive will crash within minutes of spinning up.

Practical implication for rebuilders: A hard disk that has not been physically opened, not been dropped, and has failed only electronically (failed controller board, failed motor driver) is often repairable by replacing the faulty electronics. A drive that has experienced a head crash or been opened in a non-cleanroom environment is usually unrecoverable.

The Actuator

The actuator is the mechanism that moves the heads radially across the platter surfaces, seeking to different tracks. Two main designs exist:

Stepper motor actuator (older drives, floppy drives): A stepper motor moves the head carriage in discrete steps, one track per step. Positioning is open-loop — the controller assumes the head is where it commanded, with no position feedback from the disk. Simpler electronics, but slower, less accurate on worn drives, and sensitive to track registration errors.

Voice coil actuator (all modern hard drives): A permanent magnet and a coil-wound arm form a voice coil motor (essentially a linear version of a loudspeaker). Current through the coil produces force that rotates the arm. Position feedback comes from servo data written on the disk surface — the head reads its own position and the electronics use this to correct the arm position continuously (closed-loop control). Voice coil actuators are faster, more accurate, and wear-free compared to stepper motors.

Seek time: The time for the actuator to move from one track to another. Average seek time (average across random track pairs) is typically 5–15 milliseconds in modern drives. Combined with average rotational latency (half a revolution: ~4 ms at 7,200 RPM), average access time is around 10–20 milliseconds.

Data Organization and File Systems

Raw disk capacity is organized by the operating system into a usable structure through a file system. The disk itself only knows about cylinders, tracks, and sectors. The file system maps filenames and logical files to specific sectors.

Partition table: The very first sector of the disk (the master boot record, MBR) contains a partition table describing how the disk is divided into regions. Each partition is an independent region that can hold a separate file system.

Boot sector: The first sector of a partition contains code that the computer executes when booting (starting up), plus file system metadata: total sectors, sector size, number of FAT/inode tables, root directory location.

File Allocation Table (FAT): A simple file system structure that maps each data block to the next block in the same file (or marks it as the last block). Finding a file requires following a linked list of block pointers. Simple to implement, but fragmentation over time degrades performance.

Unix inode structure: Each file has a fixed-size data structure (inode) containing file metadata and pointers to data blocks. Direct pointers (for small files), indirect pointers (for medium files), and double/triple indirect pointers (for large files). More complex but avoids fragmentation and scales to large files.

For a minimal rebuild system, FAT is the easiest file system to implement from scratch. FAT was used in CP/M, MS-DOS, and early Windows, and its format is fully documented and simple enough to implement in a few hundred lines of code.

Reliability and Failure Modes

Hard disks fail in several ways:

Head crash: Physical contact between head and disk, causing immediate catastrophic failure. Prevented by: keeping the drive horizontal during operation (early drives required this), avoiding shock and vibration, maintaining sealed enclosure.

Bad sectors: Individual sectors develop read errors due to local magnetic coating defects, particle contamination, or thermal calibration errors. Modern drives maintain a spare sector pool and transparently remap bad sectors to spares. A drive that begins accumulating bad sectors rapidly is failing and should be replaced immediately.

Motor bearing failure: The spindle bearings wear over time. Symptoms: increased noise (grinding, clicking), occasional read errors, eventual failure to spin up. Not repairable in the field.

Controller board failure: The PCB attached to the bottom of the drive contains motor driver, read/write channel, and interface electronics. It can fail due to power surges or component aging. For some drive models, swapping the controller board from an identical drive recovers the data (data is not stored on the controller board).

Thermal calibration drift: Drive heads expand and contract slightly with temperature. Modern drives periodically recalibrate head position against servo data. If this calibration drifts, read errors appear. Usually self-correcting; if persistent, indicative of a failing drive.

S.M.A.R.T. monitoring: All modern drives self-monitor dozens of health parameters (reallocated sector count, spin-up time, error rates) and report them via the S.M.A.R.T. interface. Monitoring these parameters gives advance warning of impending failure in most cases.

Practical Use in a Rebuilt System

A salvaged hard disk drive requires only a working interface controller (ATA/IDE is the simplest and most common) and a power supply (+5V and +12V DC). The drive does not need a computer to function — it can be operated with a minimal controller that issues simple commands:

Basic ATA operations: IDENTIFY DRIVE (get capacity and parameters), READ SECTORS (specify cylinder/head/sector address, read N sectors into buffer), WRITE SECTORS (write N sectors from buffer to specified address).

These three operations are sufficient to implement a functional storage system. The ATA command set has been implemented on simple microcontrollers and even in FPGAs with as little as a few kilobytes of code.

For maximum longevity, avoid unnecessary seeks (organize data to minimize head movement), implement write buffering (accumulate writes into full-track or full-cylinder writes), and monitor S.M.A.R.T. data weekly. Keep at least two copies of critical data on different physical drives.