Disk Structure & RAID Levels

Section 01

The Story That Explains Disk Structure & RAID

📖 Real World Analogy

The Warehouse of Rotating Shelves

Imagine a book warehouse where books live on rotating circular shelves stacked vertically. A single robotic arm can slide up-and-down to reach any shelf level, and each shelf spins to bring the right book past the arm. The warehouse manager has to solve three separate problems:

Structure — how are books physically organised on the shelves? By subject, alphabetical, or radial position?
Formatting — before any books arrive, we must pre-print the labels on every shelf slot so the arm knows where to place and find each book.
Redundancy — what happens if a shelf catches fire? A responsible manager doesn't keep one copy of every book. She uses mirror shelves or clever error-correction codes so a single failure doesn't lose data.

These three concerns — physical geometry, formatting, and fault tolerance — are the same ones an OS solves for disk drives. Physical layout is disk structure. Pre-printing the labels is disk formatting. Multiple redundant copies is RAID.

💡

Why It All Matters Together

A file system sits on top of a formatted disk. That formatted disk may itself sit on top of a RAID array of many physical disks. Understanding all three layers is essential for any system administrator, backend engineer, or storage architect.

Section 02

Physical Disk Structure

Anatomy of a Hard Disk

A hard disk contains one or more platters rotating on a common spindle. Each platter has two magnetic surfaces, each served by one read/write head. All heads move together on a single actuator arm.

🛠️ Geometry Vocabulary

Platter

A physical disc coated with magnetic material. 3–8 typical in a modern HDD.

Surface

One side of a platter (each platter has two — top and bottom).

Head

Read/write device flying above one surface. One head per surface.

Track

Circular path at a fixed radius on a surface.

Sector

Fixed-size arc (usually 512 B or 4 KB) — smallest addressable unit.

Cylinder

All tracks at the same radius across every platter. Reading a whole cylinder needs no arm movement.

Cluster

File-system-level grouping of several sectors, used for allocation.

Section 03

Disk Addressing — CHS vs LBA

Every sector on the disk needs a unique address. Two schemes exist historically.

🏠 CHS — Cylinder / Head / Sector

Legacy three-part address: (C, H, S)

Reflects the physical geometry

Limited by BIOS: max ~8 GB on old systems

Sector numbering starts at 1, cylinder/head at 0

🔢 LBA — Logical Block Addressing

Single integer: LBA 0, 1, 2, …

OS-friendly, hides physical geometry

Modern systems use 48-bit or 64-bit LBAs

The disk controller translates LBA → internal geometry

Conversion Formulas

// Given: H heads/cylinder, S sectors/track
// CHS (c, h, s) → LBA
LBA = (c × H + h) × S + (s − 1)

// LBA → CHS
c = LBA / (H × S)
h = (LBA / S) mod H
s = (LBA mod S) + 1

🎮 Interactive — CHS ↔ LBA Conversion

Disk geometry: H = 4 heads, S = 20 sectors/track. Convert CHS (5, 2, 15) to LBA, then convert LBA 454 back to CHS.

CHS ↔ LBA conversion — step through the math

Given

          H = 4, S = 20 sectors/track
        

Result

—

Computation Trace

Click Next to begin

Step 0 of 8

START Two conversions to walk through: (1) CHS (5, 2, 15) → LBA, then (2) LBA 454 → CHS. Click Next.

Section 04

Disk Formatting

Before a disk can store user files, it must be formatted. This actually happens in two stages, at very different levels.

Low-Level (Physical) Formatting

⚙️ Low-Level Format — done by manufacturer

Divide

Split each track into equal-length sectors.

Header

Write a header before each sector containing: sector number, sync pattern, and ECC.

Trailer

Write a trailer after each sector containing more ECC bits for error detection/correction.

Result

Disk is "raw formatted" — controller can now address sectors by number.

Sector Anatomy

A single sector on disk — 512 bytes of data + overhead

High-Level (Logical) Formatting

💾 High-Level Format — done by OS

Partition

Divide raw disk into one or more logical partitions.

FS Metadata

Write empty file-system data structures — FAT tables, i-node table, MFT, superblock — depending on the chosen file system.

Boot Block

Reserved first sector(s) hold bootstrap code that the BIOS/UEFI runs to start the OS.

Free-space Map

Bitmap or list marking which blocks are free.

Bad Blocks

⚠️

Sector Sparing

Every disk ships with a pool of spare sectors. When a sector goes bad, the controller silently substitutes a spare (sector remapping / sector forwarding). The OS never sees the failure — it just gets a slightly slower response. Modern SMART monitoring exposes remap counts as an early warning of impending disk death.

Section 05

RAID — Redundant Array of Independent Disks

A single disk has three big problems: it's slow, it's small, and it fails. RAID attacks all three by combining multiple physical disks into one logical unit. Different RAID levels balance the three axes: speed, capacity, and fault tolerance.

⚡

RAID 0 — Striping

speed + capacity, no safety

Split data across N disks. Read/write in parallel → N× throughput. Use all N disks' capacity. But one disk failure loses everything.

🔖

RAID 1 — Mirroring

safety + read speed

Every block written to two disks. Usable capacity halves, but survives one full disk failure. Reads can come from either copy.

🔢

RAID 4 & 5 — Parity

safety with less overhead

Store N−1 disks of data + 1 disk of XOR parity. Survives one disk failure. Uses only 1/N of raw capacity as overhead — much cheaper than mirroring for large arrays.

🔢

RAID 6

double parity

Two independent parity blocks per stripe. Survives two simultaneous disk failures. Essential for large arrays where rebuild times are long.

🔁

RAID 10 (1+0)

nested — mirror + stripe

Mirror pairs, then stripe across pairs. High performance and high fault tolerance. Popular for databases. Uses 50% of raw capacity.

🛡️

RAID 0+1 vs RAID 1+0

order matters

RAID 0+1 (stripe of mirrors) and RAID 1+0 (mirror of stripes) sound similar but have different failure modes. RAID 10 (1+0) tolerates more disk failures than 0+1.

🎮 Interactive — RAID 0 vs RAID 1

Write 4 data blocks (A, B, C, D) to a 2-disk array. Watch how RAID 0 (striping) and RAID 1 (mirroring) distribute them.

RAID 0 vs RAID 1 — striping vs mirroring

RAID 0 — Striping (2 disks)

—

RAID 1 — Mirroring (2 disks)

—

      Ready — click Next to write block A
    

Step 0 of 5

START Both arrays start empty. Four blocks A, B, C, D queued for writing. Click Next.

Section 06

RAID 5 — Distributed Parity

RAID 5 stores parity blocks alongside data blocks. For each stripe, one block holds the XOR of all other blocks in that stripe. Parity blocks rotate across disks to avoid concentrating writes on one disk (unlike RAID 4).

🔑

The XOR Magic

If P = A ⊕ B ⊕ C, then losing any one of A, B, C, or P still leaves the other three, and the missing one can be recovered by XORing them together. This is the mathematical heart of RAID 5.

🎮 Interactive — RAID 5 Parity Calculation & Recovery

RAID 5 — compute parity, then recover a failed disk

Waiting for first step…

Step 0 of 6

START 4-disk array, 3 data blocks + 1 parity per stripe. Click Next.

Section 07

Numerical Problem 1 — RAID Capacity & Fault Tolerance

Given 8 disks of 500 GB each (total raw capacity 4 TB). Compute usable capacity and maximum simultaneous disk failures tolerated for each RAID level.

RAID 0 — Striping

Usable capacity = N × C = 8 × 500 GB = 4 TB
Failures tolerated = 0   // any single failure = total loss
Efficiency = 100%

RAID 1 — Mirroring (4 mirrored pairs)

Usable capacity = N/2 × C = 4 × 500 GB = 2 TB
Failures tolerated = up to 4 (one from each pair)
Efficiency = 50%

RAID 5 — Distributed Parity

Usable capacity = (N − 1) × C = 7 × 500 GB = 3.5 TB
Failures tolerated = 1
Efficiency = 87.5%

RAID 6 — Double Parity

Usable capacity = (N − 2) × C = 6 × 500 GB = 3 TB
Failures tolerated = 2
Efficiency = 75%

RAID 10 — Mirror + Stripe (4 pairs)

Usable capacity = N/2 × C = 4 × 500 GB = 2 TB
Failures tolerated = up to 4 (one per pair)
Efficiency = 50%

RAID	Usable	Failures	Efficiency	Best For
0	4 TB	0	100%	Scratch space, video editing
1	2 TB	1–4	50%	Boot drives, small critical data
5	3.5 TB	1	87.5%	General file servers
6	3 TB	2	75%	Large arrays, long rebuilds
10	2 TB	1–4	50%	Databases, OLTP

Section 08

Numerical Problem 2 — Interactive RAID Sizing

Different problem: 12 disks of 2 TB each. Click Next to compute usable capacity for each RAID level one at a time.

RAID sizing walkthrough — 12 × 2 TB

Given

          N = 12 disks, C = 2 TB each

          Raw total = 24 TB

Current RAID Level

—

Calculation

        Waiting for first step…
      

Verdict

—

Step 0 of 5

START 12 disks × 2 TB = 24 TB raw. We'll compute usable capacity for each RAID level. Click Next.

Section 09

Real-World Applications

💾

Enterprise Servers

Most enterprise data centers use RAID 5 or RAID 6 for capacity-heavy workloads (file shares, backups) and RAID 10 for latency-sensitive workloads (databases, OLTP).

RAID 5 · RAID 10

🖥️

Home NAS Devices

Synology, QNAP, and TrueNAS home boxes typically default to RAID 5 or RAID 6. Modern software equivalents (Btrfs raid5, ZFS raidz1/raidz2) offer the same guarantees.

Synology · QNAP · ZFS

🌐

Cloud Object Storage

AWS S3, Azure Blob, and Google Cloud Storage use erasure coding — a generalized RAID that spreads data across many disks with configurable redundancy. Some services use 10-of-16 or 14-of-18 codes.

Erasure coding · S3

🎮

Video Editing Workstations

RAID 0 with multiple NVMe SSDs is common for 4K/8K video scratch space. Speed matters more than durability — source footage lives on backup elsewhere.

RAID 0 · NVMe

📡

Cellular Base Stations

Telecom equipment uses RAID 1 for boot drives — small, dual-mirrored SSDs so a single failure never takes a cell tower offline.

RAID 1 · Telecom

🔥

Boot Drives Everywhere

Windows Server, Linux, and VMware all support RAID 1 for OS partitions. If one drive dies, the server keeps running from the other — you replace the failed drive at leisure.

Boot mirroring

Section 10

Golden Rules — Disk Structure, Formatting & RAID

🔑 Galvin's Non-Negotiable Rules

A hard disk's physical structure — platter, surface, head, track, sector, cylinder — dictates every access-time calculation. Master this vocabulary before anything else.

Modern OSes use LBA (logical block addressing) rather than CHS. LBA hides physical geometry behind a single integer index. The disk controller translates internally.

Low-level formatting is done by the manufacturer — writes sector headers, ECC, and boot records. High-level formatting is done by the OS — writes file-system data structures onto partitions.

Every disk reserves spare sectors for automatic remapping of bad blocks. SMART tools expose the remap count as an early warning sign of impending failure.

RAID 0 stripes data for speed — capacity 100%, but zero fault tolerance. Any single disk failure loses everything. Use only for scratch data.

RAID 1 mirrors data across two disks — capacity 50%, tolerates one full failure per mirror. Ideal for boot volumes and small critical data.

RAID 5 uses distributed parity — capacity (N−1)/N, tolerates one failure. The best general-purpose choice for large file servers.

RAID 6 uses double parity — capacity (N−2)/N, tolerates two failures. Essential for large arrays where rebuild takes hours and second failures are likely.

RAID 10 (1+0) is the go-to for high-performance databases — capacity 50%, but excellent write throughput and up to N/2 failure tolerance (one per pair).

RAID is not backup. A single ransomware attack or human deletion propagates instantly to all RAID copies. Always keep offline backups separate from the array.