RAID - Redundant Array of Independent Disks

Introduction

RAID combines multiple physical disk drives into a single logical unit to achieve goals that no single disk can provide alone — higher performance, greater reliability, or both. The fundamental insight is that by spreading data across multiple disks intelligently, you can read/write faster (parallelism), survive disk failures without data loss (redundancy), or achieve a combination of both.

RAID was proposed in 1988 by researchers at UC Berkeley who observed that many cheap disks working together could outperform a single expensive disk while also being more reliable. Today, RAID is used in virtually every server, data center, and NAS (Network Attached Storage) device.

Core RAID Techniques

Striping

Distributing data across multiple disks so that reads and writes happen in parallel:

Mirroring

Writing identical data to two or more disks simultaneously. If one disk fails, the other has a complete copy.

Parity

A calculated value that allows reconstruction of lost data. If you have blocks A, B, and C, and parity P = A XOR B XOR C, then any single missing block can be recalculated from the others.

RAID Levels

RAID 0 — Striping Only (No Redundancy)

• Performance: Excellent (N× read and write speed)
• Redundancy: None — ANY disk failure loses ALL data
• Capacity: 100% (all disk space usable)
• Minimum disks: 2
• Use case: Scratch storage, video editing where speed matters and data is backed up elsewhere

Risk calculation: With N disks, probability of failure within a year ≈ N × single disk failure rate. RAID 0 with 4 disks is 4× more likely to fail than a single disk.

RAID 1 — Mirroring

• Performance: Read speed 2× (read from either disk); Write speed 1× (must write to both)
• Redundancy: Survives 1 disk failure
• Capacity: 50% (half the total disk space is usable)
• Minimum disks: 2
• Use case: Operating system disk, boot drives, critical databases requiring simple redundancy

RAID 5 — Striping with Distributed Parity

Parity is distributed across all disks (not on a dedicated disk) to avoid a parity disk bottleneck.

• Performance: Good read (N-1)×; Write slower (must calculate and write parity)
• Redundancy: Survives 1 disk failure
• Capacity: (N-1)/N — lose one disk's worth to parity (e.g., 4×1TB = 3TB usable)
• Minimum disks: 3
• Use case: General-purpose file servers, web servers — good balance of performance, capacity, and protection

Write penalty: Every write requires reading old data + old parity, calculating new parity, then writing new data + new parity. This "read-modify-write" cycle makes random writes slower.

RAID 6 — Dual Parity

Like RAID 5 but with TWO independent parity calculations (P and Q):

• Redundancy: Survives 2 simultaneous disk failures
• Capacity: (N-2)/N
• Minimum disks: 4
• Use case: Large arrays where the probability of a second failure during rebuild is significant

Why RAID 6 matters: With modern large disks (8-16TB), rebuilding a RAID 5 array after a failure takes 12-24 hours. During that rebuild, the array is vulnerable — if another disk fails, ALL data is lost. RAID 6 protects against this scenario.

RAID 10 (1+0) — Mirrors then Stripes

Combines RAID 1 (mirroring) and RAID 0 (striping):

• Performance: Excellent read and write (stripe width × mirror read advantage)
• Redundancy: Can lose one disk per mirror pair without data loss
• Capacity: 50% (same as RAID 1)
• Minimum disks: 4
• Use case: Databases (fast random writes), high-performance applications

RAID Level Comparison

Hardware RAID vs. Software RAID

Hardware RAID

• Dedicated RAID controller card with its own processor and cache
• Battery-backed write cache (protects against power loss during writes)
• OS sees a single disk — transparent
• Expensive but highest performance

Software RAID

• Implemented by the OS (Linux md, Windows Storage Spaces, ZFS)
• Uses main CPU for parity calculations
• More flexible (easy to reconfigure)
• Free (no special hardware needed)

Choosing the Right RAID Level

Ask these questions:

1. How critical is the data? (determines redundancy need)
2. What is the read/write pattern? (sequential vs random)
3. How much capacity can you sacrifice for protection?
4. How many disks are available?
5. What is the budget?

General guidelines:

• Temporary/replaceable data → RAID 0
• Critical data, simple setup → RAID 1
• File server, good balance → RAID 5
• Large array, high reliability → RAID 6
• Database, maximum performance + safety → RAID 10

Key Takeaways

• RAID uses multiple disks for performance (striping), reliability (mirroring/parity), or both
• RAID 0 is fastest but provides no protection — data loss on any disk failure
• RAID 1 mirrors data for simple redundancy at 50% capacity cost
• RAID 5 uses distributed parity for space-efficient protection but has write penalty
• RAID 6 survives two simultaneous failures (important for large arrays)
• RAID 10 combines mirroring and striping for best performance with good reliability
• RAID is NOT a backup — it protects against hardware failure, not accidental deletion or ransomware