Disk Structure

Definition

Disk Structure: Physical and logical organization of hard disk drives (HDD) and solid-state drives (SSD), including how data is organized, accessed, and managed.

HDD Physical Structure

Platter

Rotating magnetic disk:

Platter (circular magnetic surface):
┌─────────────────┐
│    Magnetic     │
│    Coating      │
│   (Iron oxide)  │
└─────────────────┘

Size: 3.5", 2.5", 1.8" diameter
Thickness: ~1.2mm
Rotating at: 5400-15000 RPM
Speed impact: Higher RPM = faster access

Spindle

Motor rotating platters:

HDD Assembly:
    ┌─────────┐
    │ Spindle │ (motor)
    └─────────┘
        ║
        ║ Rotation
        ║
    ┌─────────┐
    │Platter 1│ (5.9GB @ 15000 RPM)
    │Platter 2│
    │Platter 3│
    │Platter 4│
    └─────────┘

Multiple platters: More storage (one device)

Track

Concentric circle on platter:

Platter view (top):
    Center
       ↓
┌──────•─────────┐
│ Track 0 (innermost) (outermost)
│ Track 1         Track n
│ Track 2
│ ......
│ Track n
└────────────────┘

Tracks: Numbered from outer (0) to inner
Each track: Data storage location
Seek: Move arm to different track (expensive!)

Sector

Small arc segment of track:

Track divided into sectors:
        ┌─ Sector 0
        │  Sector 1
      ╱─┴─ Sector 2
     ╱    Sector 3
    ╱     ...
   ╱      Sector 31 (32 sectors per track typical)

Sector: Smallest addressable unit (512B or 4KB)
Block: OS unit (usually multiple sectors)

Cylinder

Vertical stack across platters:

Multiple platters, same track position:
    Platter 0: Track 5
    Platter 1: Track 5  ← Cylinder 5
    Platter 2: Track 5
    Platter 3: Track 5

Cylinder: All tracks at same radius across all platters

Access sequence:
Read cylinder 5 on all platters (no seek!)
Then move to cylinder 6

More efficient than jumping between cylinders

Disk Access Characteristics

Seek Time

Time to move arm to track:

Current Position: Track 10
Target Position: Track 50

Arm moves: 40 tracks
Speed: ~10 ms per 1000 tracks
Time: ~0.4 ms

Total: 2-5 ms typical
Fastest: 0.5 ms (minimal distance)
Worst: 10 ms (full stroke)

Rotational Latency

Wait for sector to rotate under head:

Sector needed is not under read head
Must wait for rotation

Average: Half rotation = (1 minute / RPM) / 2

Examples:
7200 RPM: 60/7200 = 8.33ms per rotation
          Average wait: 4.17ms

15000 RPM: 60/15000 = 4ms per rotation
           Average wait: 2ms

Higher RPM = Lower latency!

Transfer Time

Read sector data:

Sector size: 4KB = 4096 bytes
Transfer rate: 150 MB/s (typical modern HDD)

Time: 4096 / (150×10^6) = 27 microseconds
Negligible compared to seek + latency

Total Access Time

Total = Seek + Rotational Latency + Transfer

Example (7200 RPM HDD):
= 3ms + 4.17ms + 0.027ms
= 7.2 milliseconds

Implication:
One disk access = execute 7 million CPU instructions!
Disk I/O extremely expensive

Avoid random disk access!
Sequentially better (sequential access)

SSD Structure

No Moving Parts

Traditional HDD: Mechanical (slow, fragile)
SSD: Electronic (fast, durable)

SSD uses:
NAND Flash Memory (similar to USB drives)
No spinning, no seeking

Cell Architecture

SSD Data Storage:
┌──────────────────┐
│ Cell (bit storage)│ MLC: 2 bits per cell
│ Cell             │ TLC: 3 bits per cell
│ Cell             │ QLC: 4 bits per cell
│ Cell             │ PLC: 5 bits per cell
└──────────────────┘

More bits per cell = cheaper but slower

SSD: 256GB, 512GB, 1TB, 2TB common

Access Time

Read: ~100 microseconds (100x faster than HDD!)
Write: ~100 microseconds
Seek: None (direct electronic access)

Sequential vs Random: Minimal difference!

Why SSDs became dominant:
- 100x faster than HDD
- No mechanical latency
- Random access as fast as sequential

Logical Organization

Blocks

Minimum read/write unit from OS perspective:

Sector (physical): 512 bytes or 4KB
Block (logical): 4KB typical

Multiple sectors grouped:
Sector 0 + Sector 1 + Sector 2 + Sector 3 = Block 0

OS always reads/writes whole blocks
Never partial blocks!

File System

Layout files on disk:

Disk organized by file system:
Boot sector
File Allocation Table (FAT)
Root directory
Data blocks
...

Different file systems:
FAT32, NTFS (Windows)
ext4, XFS (Linux)
APFS (macOS)

Disk Scheduling

Why Scheduling?

Multiple I/O requests queue:

Requests in queue:
Read cylinder 50
Read cylinder 10
Read cylinder 100
Read cylinder 30

Order matters!
If serve in order: 50→10→100→30
  Seek: 40 + 90 + 70 = 200 tracks

If optimize: 10→30→50→100
  Seek: 20 + 20 + 50 = 90 tracks

2.2x better!

Disk scheduling minimizes seeking

NAND Flash Memory Considerations

Write Amplification

SSD has flash cells (blocks)

To modify data:
1. Read entire block (erase unit)
2. Update portion
3. Write back entire block

Original write: 4KB
Actual write: 64KB (worst case)
Write Amplification: 16×

Reduces SSD lifetime!

Optimization: Use levels, wear leveling

Wear Leveling

Flash cells have limited write cycles:
~10,000-1,000,000 writes per cell

Issue: If same block written repeatedly:
Cell wears out quickly

Solution: Distribute writes across all cells
SSD controller transparently moves data
Extends lifespan

User transparent (automatic)

Modern Disk Management

TRIM Command

OS tells SSD: "This data no longer used"

Example:
User deletes file (file deleted, data physical blocks freed)

HDD: Data still on disk, wasted space
SSD: TRIM tells controller to erase blocks
     Frees physical space
     Improves performance

Linux: fstrim command
macOS: Automatic on modern versions
Windows: Scheduled weekly

Garbage Collection

SSD Controller background process:
1. Identify mostly-empty blocks
2. Consolidate valid data
3. Erase empty blocks
4. Improves free space
5. Restores performance

Transparent to user
May cause latency spikes (noticed as lag)

Performance Characteristics

Disk Organization Examples

Example 1: Traditional HDD Layout

Sector 0-63: Boot code
Sector 64-511: File Allocation Table
Sector 512+: Data blocks

Data access sequence:
1. Read FAT (disk scheduling)
2. Find file's blocks in FAT
3. Schedule reads for those blocks
4. Read data in optimized order

Example 2: Modern SSD Layout

LBA (Logical Block Address): 0, 1, 2, ..., N
Maps to Physical Block Address: May be scattered

Controller maintains FTL (Flash Translation Layer)
Transparently manages:
- Wear leveling
- Garbage collection
- Error correction
- TRIM support

User sees: Single LBA space (0-N)
Actually scattered across physical cells

Summary

HDD uses mechanical seeking (slow) with tracks, sectors, cylinders. SSD uses electronic flash memory (fast) with no mechanical latency. Disk scheduling optimizes access order to minimize seeking. HDD access time dominated by mechanical latency (milliseconds). SSD access electronic (microseconds). Modern systems almost exclusively SSD. Understanding disk structure important for appreciating I/O subsystem and why disk I/O expensive. Optimization strategies differ for HDD (sequential access, scheduling) vs SSD (wear leveling, TRIM, GC).