RAID: Redundant Array of Independent Disks

What is RAID?

A single disk has two problems: it’s slow, and it can fail. RAID solves both by combining multiple disks into what looks like one — spreading data across them for speed and duplicating it for safety. RAID (Redundant Array of Independent Disks) is a disk organization technique that manages multiple disks and presents them as a single logical disk. It achieves two things simultaneously:

Speed — by using multiple disks in parallel, it delivers high capacity and high throughput.
Reliability — by storing data redundantly across disks, it ensures data can be recovered even if one disk fails.

Three Core Techniques

Mirroring

Every disk has an exact duplicate — the logical disk consists of two physical disks.

Every write is carried out on both disks simultaneously.
Reads can be served from either disk.
If one disk in a pair fails, data is still available on the other.
Data loss only occurs if both disks in a pair fail before one is repaired — extremely unlikely.

Striping

Data is split across multiple disks so reads and writes happen in parallel.

Bit-level Striping

Bits of each byte are split across multiple disks (bit \(i\) goes to disk \(i\)).
Each access reads data at up to 8× the rate of a single disk.
Seek and access time is worse than a single disk — rarely used today.

Byte-level Striping

Each file is split into 1-byte chunks distributed across disks.
The \(i\)th byte goes to disk \(((i-1) \mod n) + 1\).

Block-level Striping

Block \(i\) of a file goes to disk \((i \mod n) + 1\).
Requests for different blocks can run in parallel if they reside on different disks.
A long sequential read can utilise all disks at once.

Parity

Parity provides fault tolerance by storing extra computed information alongside the data. If a disk fails, the lost data can be reconstructed from the remaining disks and the parity — without needing a full duplicate copy.

Parity is computed using XOR: for any set of bits, XOR of all bits including the parity bit equals zero. So if one disk’s bits are lost, XOR of the remaining disks recovers them.

Bit-interleaved Parity

One parity bit covers the corresponding bits across all data disks.
Written to a dedicated parity disk on every write.
On disk failure: XOR all remaining bits (including parity) to reconstruct the lost bit.

Block-interleaved Parity

Uses block-level striping with a dedicated parity block (not just a bit) for every set of corresponding blocks across \(n\) disks.
On disk failure: XOR the corresponding blocks from all remaining disks to reconstruct the lost block.

Standard RAID Levels

RAID 0 — Block Striping

Data is striped across disks with no redundancy whatsoever.
Best write performance of all RAID levels — no parity to update.
100% storage efficiency.
No fault tolerance — one disk failure loses everything.

RAID 1 — Mirroring

Every disk has an exact copy on a second disk. Every write of a disk block involves a write on both disks.
Excellent fault tolerance — survives one disk failure.
Reads can be served from either disk in parallel.
50% storage efficiency — half your disks are copies.
Most expensive solution.

RAID 2 — Parity

Uses dedicated bit for parity and the striping is a single bit, uses Hamming codes for error correction.
Can correct one-bit errors and detect two-bit errors.
Requires multiple dedicated check disks.
Not used in practice — subsumed by RAID 3.

RAID 3 — Byte Striping + Parity

Strips at the byte level with a single dedicated parity disk.
High data transfer rate — data accessed in parallel across all disks.
Cannot service multiple requests simultaneously — every request touches every disk.
Reliability overhead for RAID 3 is a single disk, the lowest over-head possible.

RAID 4 — Block Striping + Parity

Strips at the block level with a single dedicated parity disk.
Read requests can be served by a single disk — better than RAID 3 for reads.
Can recover from one disk failure using XOR.
Write performance is poor — every write hits the single parity disk, creating a bottleneck.

RAID 5 — Distributed Parity

Like RAID 4 but parity blocks are spread across all disks instead of sitting on one dedicated disk.
Eliminates the parity bottleneck — multiple writes can happen in parallel.
All disks participate in reads — higher read parallelism.
Allows recovery of only 1 disk failure.

RAID 6 — Dual Parity

Extends RAID 5 with a second parity block, using Reed-Solomon codes.
Survives two simultaneous disk failures — safe even during a rebuild after one failure.
Write performance worse than RAID 5 — two parity blocks to update.

RAID 01 — Mirror of Stripes

Stripe sets are built first (RAID 0), then each set is mirrored (RAID 1).
If one disk in a stripe fails, the entire stripe set is lost — falls back to the mirror.
Less resilient than RAID 10.
Requires at least 4 disks. Storage efficiency: 50%.

RAID 10 — Stripe of Mirrors

Disks are mirrored first (RAID 1), then striped across the mirror pairs (RAID 0).
One disk in each mirror pair can fail independently — far more resilient than RAID 01.
Best throughput and latency of all fault-tolerant RAID levels.