An HDD (hard disk drive) is a non-volatile storage device that records data as magnetic patterns on rotating platters and reads or writes them with heads mounted on a moving actuator arm. In computer storage it remains the lowest-cost medium per terabyte for online data, while solid-state drives have taken over the role of the fast system disk.
Because every access depends on physical movement, an HDD behaves very differently from flash storage: sequential transfers are respectable, but random access is slow, and the drive is sensitive to shock, vibration and heat. Understanding these mechanical limits is the key to using hard drives well in modern hardware.
How an HDD stores and retrieves data
An HDD stores data by magnetizing tiny regions of a thin cobalt-alloy layer on aluminum or glass platters. The platters spin at a constant speed on a spindle motor, and a read/write head floats a few nanometers above the surface on an air or helium cushion created by the rotation itself.
Data is organized in concentric tracks divided into sectors, today almost always 4 KiB physical sectors (Advanced Format), often presented to the host as 512-byte logical sectors for compatibility. A voice-coil actuator swings the arm to the right track, and the drive then waits for the wanted sector to rotate under the head.
The time to reach data therefore has two mechanical parts. Seek time is the arm movement, and rotational latency is the wait for the sector, on average half a revolution. The table below shows the main components and their roles.
| Component | Function | Typical detail |
|---|---|---|
| Platters | Hold the magnetic recording layer | 1 to 10 platters, 2.5 or 3.5 inch |
| Spindle motor | Rotates the platter stack | 5,400 to 15,000 rpm |
| Heads and actuator | Position, read and write data | Two heads per platter |
| Controller and cache | Manage commands, buffering and error correction | Cache of 64 to 512 MB |
| Enclosure | Seals the mechanics from dust | Air-filled or helium-sealed |
Key performance and capacity parameters
HDD performance is defined mainly by rotational speed and areal density. Faster rotation shortens latency, while higher density puts more bits under the head per revolution and raises sequential throughput. Throughput also varies across the disk: outer tracks are faster than inner tracks because they hold more sectors per rotation.
| Parameter | Typical range | What it affects |
|---|---|---|
| Rotational speed | 5,400, 7,200, 10,000 or 15,000 rpm | Latency and throughput |
| Average rotational latency | About 2 to 5.6 ms | Random access time |
| Average seek time | About 4 to 12 ms | Random access time |
| Sequential transfer | About 100 to 300 MB/s | Large file copies and backups |
| Random 4 KiB operations | About 75 to 200 IOPS | Databases and many small files |
| Capacity | 1 TB to more than 30 TB | Cost per terabyte |
Recording technology also matters. CMR (conventional magnetic recording) writes tracks side by side. SMR (shingled magnetic recording) overlaps tracks like roof tiles to gain capacity, but rewriting data forces whole zones to be rewritten, which causes severe slowdowns during sustained random writes. Energy-assisted methods such as HAMR heat the medium briefly with a laser to allow even smaller, stable magnetic grains.
HDD compared with SSD and other storage
The essential difference between an HDD and an SSD is mechanical versus electronic access. An SSD stores data in NAND flash and reaches any block in tens of microseconds, so its random performance is hundreds to thousands of times higher. An HDD, in turn, offers lower cost per terabyte and does not wear out from write cycles in the way that TLC or QLC flash does.
| Property | HDD | SATA SSD | NVMe SSD |
|---|---|---|---|
| Access latency | Several milliseconds | Below 0.1 ms | Below 0.1 ms |
| Sequential speed | 100 to 300 MB/s | Up to about 550 MB/s | About 2 to 14 GB/s |
| Random IOPS | Hundreds | Tens of thousands | Hundreds of thousands or more |
| Cost per terabyte | Lowest | Higher | Higher |
| Shock sensitivity | High while operating | Low | Low |
| Write endurance | Not limited by TBW in the same way | Rated in TBW | Rated in TBW |
Most HDDs connect over SATA, which caps a link at 6 Gbit/s, still more than a single hard drive can saturate. Enterprise models often use SAS for dual-port connections and better error handling. NVMe hard drives exist in specifications but remain rare.
Choosing and using an HDD
Hard drives are the right choice where capacity matters more than latency: backups, media libraries, surveillance recording, archives and large network storage arrays. They are a poor choice for the operating system disk or for workloads dominated by small random reads.
| Workload | Recommended drive | Reason |
|---|---|---|
| Desktop bulk storage | 7,200 rpm CMR desktop drive | Good throughput at low cost |
| NAS with RAID | NAS-rated CMR drive | Vibration tolerance and fast error recovery |
| Video surveillance | Surveillance-rated drive | Optimized for continuous sequential writes |
| Cold archive | High-capacity SMR or CMR drive | Lowest cost per terabyte, data rarely rewritten |
| System and applications | SSD instead of HDD | Random access dominates boot and launch times |
Drives intended for arrays support time-limited error recovery, so a hard-to-read sector does not stall the whole RAID set. It is also worth checking the workload rating in terabytes per year and the warranty length, because they reveal the duty cycle the drive was designed for.
Failure modes and common mistakes
HDD failures are often gradual and detectable. The S.M.A.R.T. self-monitoring system records counters such as reallocated sectors, pending sectors and read error rates, and a rising count of reallocated or pending sectors is a strong warning sign. Mechanical failures, including a head crash, can still happen suddenly, especially after a drop while running.
- Using SMR drives in RAID rebuilds can make reconstruction extremely slow or cause drives to drop out of the array.
- Moving or knocking a running drive risks head contact with the platter surface.
- Treating RAID as a backup ignores deletion, ransomware and controller failure; a separate copy is still needed.
- Ignoring temperature shortens lifespan; most drives are specified for about 5 to 60 °C during operation.
- Defragmenting an SSD by habit is unnecessary; defragmentation helps only HDDs, where scattered fragments cost extra seeks.
The idea in brief
A hard drive works a little like a record player that can also record. Inside, metal disks spin very fast. A tiny arm moves across them to read or write information stored as magnetism.
The arm has to move, and the disk has to turn to the right spot. That makes a hard drive slower than an SSD when it must jump between many small pieces of data. Its big advantage is price. It stores large amounts of data cheaply. That makes it ideal for photos, videos and backups, but not for starting the computer.
A practical scenario
A small architecture office keeps project files on a computer with a 1 TB NVMe SSD, but its archive of drawings, renderings and scans has grown to 18 TB. Buying that much flash storage would be expensive, so the office installs a four-bay network storage device with four 12 TB NAS-rated CMR hard drives in a RAID 5 array, giving about 36 TB of usable space.
Daily work still happens on the SSDs, while finished projects move to the array, where large sequential reads of 200 MB/s per drive are more than fast enough over the network. The administrator checks S.M.A.R.T. reports monthly and keeps a second copy on external drives stored off site. When one drive starts reporting pending sectors, it is replaced before failure, and the array rebuilds without data loss.