Disk Structure: Understanding Physical Storage

Modern hard disk drives achieve high storage capacities by stacking multiple platters on a single spindle, each providing two recording surfaces. Understanding how these multiple surfaces work together is essential for comprehending disk addressing, access patterns, and database storage optimization.

Why Multiple Platters Matter for Databases:

The way data is distributed across multiple platter surfaces directly impacts:

• Access time — Whether accessing data requires head switching (fast) or seeks (slow)
• Data layout optimization — How database systems organize data to minimize expensive operations
• Parallel access — Can multiple heads read simultaneously?
• Reliability — More components mean more potential failure points

In this page, we examine the intricate relationship between platters and heads, introducing the critical concept of the cylinder—a vertical stack of tracks that can be accessed without moving the actuator arm.

Modern hard disk drives stack multiple platters on a common spindle, allowing capacity to scale without increasing the drive's footprint. Each platter provides two recording surfaces (top and bottom), effectively multiplying storage capacity.

The Stack Configuration:

A typical enterprise drive might have the following configuration:

• 9 platters stacked with precise spacing
• 18 recording surfaces (2 per platter)
• 18 read/write heads (1 per surface)
• All platters rotate together at identical speed
• All heads move together on a single actuator assembly

Critical Design Constraint:

All heads in the stack are mechanically linked to the same actuator arm assembly. This means:

• When the actuator moves, all heads move together
• You cannot position heads independently over different tracks
• At any given moment, all heads are positioned over the same track number on their respective surfaces

Surface Numbering Convention:

Recording surfaces are typically numbered sequentially:

• Surface 0: Top of first (topmost) platter
• Surface 1: Bottom of first platter
• Surface 2: Top of second platter
• Surface 3: Bottom of second platter
• ... and so on

Why Not Use All Surfaces?

Some configurations deliberately leave surfaces unused:

• Outermost surfaces may be dedicated to servo data
• Some surfaces may be reserved as spares
• Manufacturing defects may disable specific surfaces
• Firmware may disable surfaces for product segmentation

Each read/write head is mounted on a slider—a precision-engineered component that maintains the exact flying height above the platter surface. Understanding slider technology is essential for appreciating the engineering marvel that enables modern storage densities.

Air Bearing Principles:

The slider uses aerodynamic principles to 'fly' above the platter:

1. The spinning platter creates a thin layer of moving air on its surface
2. The slider's air bearing surface (ABS) is precisely shaped to generate lift
3. A spring-loaded suspension (flexure) applies a constant loading force
4. Equilibrium is reached where air pressure equals spring force

The result is a stable flying height of approximately 3-5 nanometers above the platter surface.

The Suspension System:

The slider is mounted on a thin metal flexure that provides:

• Pitch and Roll Compliance — Allows slider to follow platter curvature
• Gram Load Force — Typically 1.5-3 grams pressing toward the platter
• Electrical Connections — Traces carry signals to/from the head
• Controlled Stiffness — Optimized for resonance characteristics

Thermal Fly-Height Control (TFC):

Modern drives use a heating element embedded in the slider to dynamically adjust flying height:

1. Current through a resistive heater causes local thermal expansion
2. The read/write element protrudes slightly closer to the platter
3. Controller adjusts heater power based on read signal quality
4. Enables sub-nanometer flying height control

This allows drives to maintain optimal head-to-disk spacing across different operating conditions and platter regions.

The cylinder is one of the most important concepts in disk storage, providing the foundation for understanding efficient data access. A cylinder represents all tracks that can be accessed without moving the actuator arm.

Definition:

A cylinder is the set of tracks at the same radial position across all recording surfaces. Since all heads move together, when the actuator is positioned at track N, every head is simultaneously over track N on its respective surface.

Visualizing Cylinders:

Imagine looking down through a multi-platter stack:

• Each platter has concentric circular tracks
• Tracks at the same radius on all platters form a vertical 'cylinder'
• There are as many cylinders as there are tracks per surface
• All data within one cylinder can be accessed by head switching alone

Why Cylinders Matter:

The cylinder concept is fundamental because:

1. Accessing different sectors within a cylinder requires only head switching — An electronic operation taking microseconds
2. Accessing sectors on different cylinders requires a seek — A mechanical operation taking milliseconds

This massive difference (1,000x or more) makes cylinder-aware data layout critically important.

Cylinder Count in Modern Drives:

A modern high-capacity drive might have:

• 500,000+ cylinders per surface
• Each cylinder spanning 18 surfaces
• Hundreds of sectors per track
• Total capacity = cylinders × surfaces × sectors per track × bytes per sector

Understanding the dramatic performance difference between head switching and seeking is crucial for database storage optimization. These two operations represent fundamentally different categories of latency.

Head Switching:

When the controller activates a different head (switches surfaces) without moving the actuator arm:

• Mechanism: Electronic signal routing change
• Time: Approximately 1-2 milliseconds (often hidden in rotational latency)
• Energy: Negligible
• Wear: None

Seeking:

When the actuator arm must move to position heads over a different track:

• Mechanism: Voice coil motor accelerates/decelerates the arm assembly
• Time: 3-15+ milliseconds depending on distance
• Energy: Proportional to acceleration required
• Wear: Minimal but non-zero

The Cylinder-First Access Pattern:

To maximize performance, data should be accessed in a pattern that exhausts each cylinder before moving to the next:

1. Read/write all sectors on the current track
2. Switch to the next head (same cylinder, different surface)
3. Read/write all sectors on that track
4. Repeat until all surfaces in the cylinder are processed
5. Only then seek to the next cylinder

This pattern minimizes the number of expensive seek operations.

Modern Complication: Zone Bit Recording

In modern drives with Zone Bit Recording (ZBR), different tracks have different numbers of sectors. This complicates the cylinder concept because the 'shape' of a cylinder is no longer uniform—outer tracks hold more data than inner tracks.

Each recording surface on a platter is a precisely engineered substrate capable of storing magnetic patterns with nanometer-scale features. Understanding surface organization helps explain storage capacity and access patterns.

Surface Organization:

Each surface is divided into:

• Tracks — Concentric circles of data (50,000-500,000+ per surface)
• Sectors — Arc-shaped subdivisions of tracks (typically 512 bytes or 4096 bytes each)
• Zones — Groups of tracks with the same sector count (for Zone Bit Recording)
• Servo Wedges — Interleaved positioning information

The Data Recording Process:

Writing:

1. Write current flows through the thin-film inductive element
2. Magnetic field is generated, oriented perpendicular to the platter surface (PMR)
3. Magnetic domains in the recording layer flip to match the field orientation
4. Domain orientation persists after write current stops (non-volatile storage)

Reading:

1. Platter rotation moves recorded domains under the read head
2. Magnetic field from domains affects resistance of the GMR/TMR sensor
3. Resistance changes create voltage variations (microvolts)
4. Signal processing recovers the original bit stream

Error Detection and Correction:

The raw signal from the head contains significant noise and timing jitter. The read channel uses sophisticated digital signal processing:

• Timing recovery circuits synchronize with the bit stream
• Viterbi detection handles inter-symbol interference
• LDPC or similar codes correct bit errors
• CRC codes detect any remaining uncorrected errors

A natural question arises: Can multiple heads read or write simultaneously? The answer has significant implications for potential parallelism in disk access.

The Reality:

Conventional HDDs: One Active Head at a Time

Most hard disk drives activate only one read/write head at any moment:

• Technical Reason: The read preamplifier and write driver are shared resources
• Cost Reason: Separate electronics for each head would increase expense
• Timing Reason: Synchronizing multiple bit streams would be complex

When switching heads, there is a brief delay ( ~1-2 ms) to reconfigure the electronics and wait for the correct sector to rotate under the new head.

Does This Limit Performance?

The single-active-head constraint is less limiting than it might appear:

• Head switching overlaps with rotational latency
• Sequential access patterns naturally use heads efficiently
• The bottleneck is usually seek time, not head switching

Dual-Actuator Technology:

Recent enterprise drives have introduced dual-actuator designs:

• Two independent actuator arms in a single drive
• Each actuator controls half of the heads
• Effectively doubles IOPS (I/O Operations Per Second) for random workloads
• Presents as a single logical drive or two independent drives

Example: Seagate MACH.2

Seagate's multi-actuator technology:

• Top actuator: Accesses upper platters
• Bottom actuator: Accesses lower platters
• Each actuator can seek independently
• Random read IOPS nearly doubled compared to single-actuator
• Particularly beneficial for cloud and hyperscale data centers

Understanding failure modes is essential for database reliability engineering. The heads and platters are the most mechanically sensitive components, and their failure patterns inform backup and redundancy strategies.

Head Crash:

The most catastrophic failure mode occurs when the read/write head contacts the spinning platter surface:

Causes:

• Physical shock or vibration
• Power failure during operation (head doesn't park properly)
• Contamination particles
• Component wear over time

Consequences:

• Head and media damage, often spreading across the platter
• Complete data loss on affected surfaces
• Particulate debris may damage other heads and surfaces

Prevention:

• Shock sensors that park heads on impact detection
• Automatic head parking on power loss (ramp unload)
• Controlled operating environment (temperature, vibration)

S.M.A.R.T. Monitoring:

Self-Monitoring, Analysis, and Reporting Technology provides early warning of potential failures:

Key Indicators for Head/Platter Health:

• Reallocated Sector Count — Sectors remapped due to read errors; increasing count indicates media degradation
• Current Pending Sector Count — Sectors suspected bad awaiting rewrite
• Uncorrectable Sector Count — Sectors that failed even after ECC
• Spin Retry Count — Failed attempts to spin up; may indicate motor or bearing issues
• Raw Read Error Rate — Frequency of read errors before correction
• Seek Error Rate — Failures in head positioning

Database Administrator Best Practice:

Monitor S.M.A.R.T. data for all drives storing database data. Set alerting thresholds for critical attributes. Replace drives proactively when indicators suggest impending failure. Maintain redundancy (RAID) to survive individual drive failures.

Understanding how disk capacity is calculated from physical parameters provides insight into drive specifications and helps verify vendor claims.

The Capacity Formula:

Capacity = Cylinders × Heads × Sectors per Track × Bytes per Sector

However, due to Zone Bit Recording (ZBR), this classic formula is now approximate because sectors per track varies across zones.

Modern Capacity Calculation:

Capacity = Total Number of Logical Sectors × Sector Size

Where:

• Total sectors = sum of sectors across all zones on all surfaces
• Sector size = 512 bytes (legacy) or 4096 bytes (Advanced Format)

Capacity vs Usable Space:

The raw capacity reported by drive manufacturers differs from what operating systems show:

Drive Manufacturers Use:

• 1 TB = 10¹² bytes = 1,000,000,000,000 bytes (decimal)

Operating Systems Use:

• 1 TiB = 2⁴⁰ bytes = 1,099,511,627,776 bytes (binary)

Result:

• An '18 TB' drive has 18 × 10¹² bytes = 18,000,000,000,000 bytes
• In binary: 18,000,000,000,000 / 1,099,511,627,776 = 16.37 TiB
• The OS might report approximately 16.4 TB using hybrid units

Additional Capacity Overhead:

• Formatting overhead — File system metadata consumes some space
• Reserved blocks — Some space reserved for spare sectors and drive operations
• Partition alignment — Alignment to sector boundaries may consume space

We have completed a comprehensive examination of how platters and heads work together in multi-surface disk configurations. Let's consolidate the key concepts:

What's Next:

With this understanding of multi-platter architecture and the cylinder concept, we will now examine Tracks and Sectors in detail—how the recording surface is organized into addressable units, and how Zone Bit Recording maximizes capacity by varying sector counts based on track radius.