9.1 Magnetic Disk Anatomy: Platters, Tracks, Sectors, Seek Time & Rotational Latency

Back to Operating Systems Notes

📚Module 09: Storage & Disk SchedulingTopic 9.1⏱️20 min read

🎯High-Yield For:Computer Science Foundations • Systems Engineering • Storage Architecture

💡 Core Intuition

🍳 The Everyday Analogy: The Vinyl Record Turntable

Imagine a high-fidelity vintage vinyl record player:

Architecture Flow

The Vinyl Turntable Access Pipeline

Mapping mechanical audio playback to magnetic hard disk operations

💡 Hover or click any card for deep-dive operational details

🦾The Arm Move

Moving Tone Arm to Song Groove

Seek Time (Mechanical Arm)

You lift the turntable tone arm and slide it horizontally across the record to Track 4.

→

Wait For Groove

🔄The Spin

Waiting for Song Intro Under Needle

Rotational Latency

The needle hovers over Track 4, waiting for the platter to rotate the beginning of the song under the stylus.

→

Audio Streaming

🎵Sound Playback

Reading Music Vibrations

Transfer Time

The needle vibrates against the spinning groove, converting bumps into audio signals.

• Seek Time: Mechanically moving the physical actuator arm to the correct radial track.
• Rotational Latency: Waiting for the platter to spin the desired sector beneath the read/write head.
• Transfer Time: Electromagnetically streaming the binary bits off the platter surface into the disk controller buffer.

---

💻 Bridging to Computer Science

Secondary storage holds the vast majority of persistent data in computing systems. Unlike solid-state DRAM chips, Hard Disk Drives (HDDs) are electro-mechanical storage devices whose performance is fundamentally constrained by mechanical physics (motors, inertia, magnetic induction, and rotational drag).

The operating system's primary storage objective is two-fold:

1. Minimize Access Latency: Reducing the time between an I/O request and data delivery.
2. Maximize Disk Bandwidth: Maximizing the rate of useful data transferred per unit time: $\text{Disk Bandwidth} = \frac{\text{Total Bytes Transferred}}{\text{Total Time Between Request Arrival and Transfer Completion}}$

---

📚 Core Deep-Dive & Concepts

1. Mechanical Anatomy of a Magnetic Disk

Hard Disk Drive (HDD) Physical Platter Assembly

Multi-platter spindle stack showing mechanical actuator and head alignment

Top Assembly

Spindle Motor & Spindle Shaft

↓

Platter 1 (Surfaces 0 & 1)

Top Platter & Dual Read/Write Heads

↓

Platter 2 (Surfaces 2 & 3)

Middle Platter & Dual Read/Write Heads

↓

Platter 3 (Surfaces 4 & 5)

Bottom Platter & Dual Read/Write Heads

↓

Actuator Assembly

Voice-Coil Motor (VCM) & Actuator Arm

---

2. The Head Actuator & Cylinder Geometry

• Synchronous Head Movement: All read/write heads are fixed to a single rigid Actuator Arm Assembly. When the actuator motor fires, all heads move inward or outward together across the platters in lockstep.
• The Cylinder Concept: Because all heads hover over the same radial track simultaneously, switching between tracks within the same cylinder requires zero mechanical arm movement—only an electronic head selection switch ($< 1\,\mu\text{s}$)!
• Disk Addressing Hierarchy: To read a specific byte, the controller uses the geometric tuple: $\mathbf{\text{Disk Address} = [\,\text{Cylinder (Track)} \;\longrightarrow\; \text{Surface (Head)} \;\longrightarrow\; \text{Sector} \;\longrightarrow\; \text{Byte}\,]}$

---

3. Total Disk Access Time: Mathematical Formulation

The total latency required to service a disk I/O request is decomposed into four sequential phases:

$\mathbf{\text{Total Access Time} = \text{Seek Time} + \text{Rotational Latency} + \text{Transfer Time} + \text{Controller Overhead}}$

Disk I/O Access Latency Breakdown

Sequential hardware stages incurred from request arrival to RAM delivery

1

→

Phase 1: Controller Overhead

Host controller parses I/O command, translates Logical Block Address (LBA) to physical cylinder/head/sector, and signals hardware bus.

2

→

Phase 2: Seek Time (Dominant Bottleneck)

Mechanical arm accelerates, travels across platters, and settles read/write heads precisely over target cylinder.

3

→

Phase 3: Rotational Latency

Head hovers over track while spindle rotates target sector directly beneath the head (half-revolution average).

4

→

Phase 4: Transfer Time

Head reads magnetic flux transitions, decodes bits via ECC, and streams data into drive buffer and host RAM.

A. Seek Time ($T_{\text{seek}}$)

• The mechanical time required for the actuator arm to accelerate, travel across platters, and settle the read/write head precisely over the target track.
• Typically ranges from $3\text{ ms}$ (track-to-track) to $10\text{ ms}$ (full-stroke). Average seek time is typically $4\text{–}8\text{ ms}$.
• The Dominant Bottleneck: Seek time accounts for over $70\%$ of total random I/O latency on magnetic disks!

B. Rotational Latency ($T_{\text{rot}}$)

• The time spent waiting for the target sector on the spinning platter to arrive directly beneath the head.
• For a disk spinning at $R$ Revolutions Per Minute (RPM): $\text{Time for One Complete Revolution } (T_{\text{rev}}) = \frac{60}{R} \text{ seconds} = \frac{60 \times 1000}{R} \text{ milliseconds}$
• Because the target sector can be positioned anywhere when the head arrives, on average it must travel half a revolution: $\mathbf{\text{Average Rotational Latency } (T_{\text{rot}}) = \frac{1}{2} \cdot T_{\text{rev}} = \frac{30 \times 1000}{R} \text{ ms}}$

Spindle Speed (RPM)	Single Revolution Time ($T_{\text{rev}}$)	Average Rotational Latency ($T_{\text{rot}}$)
5400 RPM	$11.11\text{ ms}$	$\mathbf{5.56\text{ ms}}$
7200 RPM	$8.33\text{ ms}$	$\mathbf{4.17\text{ ms}}$
10,000 RPM	$6.00\text{ ms}$	$\mathbf{3.00\text{ ms}}$
15,000 RPM	$4.00\text{ ms}$	$\mathbf{2.00\text{ ms}}$

C. Transfer Time ($T_{\text{trans}}$)

• The time required to stream the binary data off the spinning media once the sector arrives under the head: $\mathbf{T_{\text{trans}} = \left( \frac{\text{Data to Transfer}}{\text{Track Capacity}} \right) \times T_{\text{rev}} = \left( \frac{\text{Data to Transfer}}{\text{Transfer Rate}} \right)}$

---

4. Rigorous Worked Numerical Problems

Problem 1: Disk Geometry & Capacity Sizing

• Problem: Consider a disk system with $512$ tracks per surface, where each track contains $128$ sectors, and each sector stores $256$ bytes.

  1. Calculate the total storage capacity of one surface.
  2. Determine the number of bits required in a disk address to specify the track, sector, and byte.

• Solution:

  1. Capacity Calculation: $\text{Tracks} = 512 = 2^9$ $\text{Sectors per track} = 128 = 2^7$ $\text{Bytes per sector} = 256 = 2^8$ $\text{Surface Capacity} = 2^9 \times 2^7 \times 2^8\text{ bytes} = 2^{24}\text{ bytes} = \mathbf{16\text{ MB}}$
  2. Address Bit Decomposition:
     • Track bits $= \log_2(512) = \mathbf{9\text{ bits}}$
     • Sector bits $= \log_2(128) = \mathbf{7\text{ bits}}$
     • Byte offset bits $= \log_2(256) = \mathbf{8\text{ bits}}$
     • Total bits $= 9 + 7 + 8 = \mathbf{24\text{ bits}}$ (which addresses $2^{24} = 16\text{ MB}$ directly).

---

Problem 2: Full End-to-End Transfer Time Calculation

• Problem: A hard disk rotates at $1500\text{ RPM}$. Each track contains $400$ sectors, and each sector stores $512$ bytes. There are $1000$ tracks on the disk. An application requests a contiguous file of size $1\text{ MB}$ ($2^{20}\text{ B} = 1,048,576\text{ bytes}$). The average seek time is $4\text{ ms}$. Calculate the total time required to transfer the entire file.

• Solution:

  1. Compute Revolution Time ($T_{\text{rev}}$): $\text{Spindle Speed} = 1500\text{ RPM} \implies \frac{1500\text{ rev}}{60\text{ sec}} = 25\text{ rev/sec}$ $T_{\text{rev}} = \frac{1}{25}\text{ sec} = \frac{1000\text{ ms}}{25} = \mathbf{40\text{ ms}}$
  2. Compute Average Rotational Latency ($T_{\text{rot}}$): $T_{\text{rot}} = \frac{T_{\text{rev}}}{2} = \frac{40\text{ ms}}{2} = \mathbf{20\text{ ms}}$
  3. Compute Track Capacity: $\text{Track Size} = 400 \times 512\text{ bytes} = 204,800\text{ bytes}$
  4. Compute Transfer Time ($T_{\text{trans}}$): $T_{\text{trans}} = \left( \frac{\text{File Size}}{\text{Track Size}} \right) \times T_{\text{rev}} = \left( \frac{1,048,576}{204,800} \right) \times 40\text{ ms} = 5.12 \times 40\text{ ms} = \mathbf{204.8\text{ ms}}$
  5. Total Access Time: $\text{Total Time} = T_{\text{seek}} + T_{\text{rot}} + T_{\text{trans}} = 4\text{ ms} + 20\text{ ms} + 204.8\text{ ms} = \mathbf{228.8\text{ ms}}$

---

5. Contiguous vs Non-Contiguous File Allocation Latency Impact

Why do operating systems aggressively defragment mechanical hard drives?

• Scenario: Consider a file requiring $8$ sectors to store. Disk parameters: Seek Time $= 15\text{ ms}$, Rotational Speed $= 3000\text{ RPM}$ ($T_{\text{rev}} = 20\text{ ms}$, $T_{\text{rot}} = 10\text{ ms}$). Track capacity $= 8$ sectors.

---

6. Sector Interleaving

In early disk controllers, after reading a sector, the hardware required a brief computational window ($0.5\text{ ms}$) to process error-correcting codes (ECC) and buffer the data before it could read the next sector:

• If sectors were numbered sequentially ($0, 1, 2, 3\dots$), by the time the controller finished processing sector $0$, sector $1$ had already spun past the head! The drive had to wait an entire revolution just to read sector $1$.
• The Solution: Interleaving. Numbering physical sectors with intentional angular gaps:
  • No Interleaving: 0, 1, 2, 3, 4, 5, 6, 7 (requires 8 full revolutions to read a track if controller is slow).
  • Single Interleaving (1:2): 0, 4, 1, 5, 2, 6, 3, 7 (reads entire track in 2 revolutions).
  • Modern drives have fast on-disk DRAM caches that eliminate interleaving entirely.

---

🏭 In The Real World: Production Case Study

High-Density Datacenter Storage: SMR & NVMe Flash

Modern cloud datacenters (AWS S3, Google Cloud Storage) manage petabytes of storage by balancing mechanical drives and flash media:

Cloud Datacenter Storage Tiering Architecture

Balancing latency, bandwidth, and cost-per-gigabyte in modern cloud infrastructure

Hot Tier

NVMe PCIe 5.0 Flash SSDs

↓

Warm Tier

Conventional PMR Hard Drives (7200 RPM)

↓

Cold / Archive Tier

Shingled Magnetic Recording (SMR) 28 TB HDDs

1. Why HDDs Still Dominate Cloud Capacity:
   • Despite flash SSDs, HDDs provide $5\times$ lower cost-per-gigabyte than NVMe flash.
2. Shingled Magnetic Recording (SMR):
   • Overlaps magnetic tracks like shingles on a roof to pack up to $28\text{ TB}$ per drive.
   • Because writing one track overwrites its neighbor, the OS kernel must write sequentially, making intelligent Disk Scheduling vital.

---

🎯 Exam & Interview Pitfall Check

Core Conceptual Questions

Question 1: Why is average rotational latency defined as half the revolution time ($\frac{1}{2} T_{\text{rev}}$)? Answer:

1. When the read/write head arrives at the target cylinder, the target sector can be at any angular position along the circumference with uniform probability:
   • Best-case scenario: The target sector happens to be directly under the head ($T_{\text{rot}} = 0$).
   • Worst-case scenario: The target sector just passed under the head a microsecond earlier, requiring a full $360^\circ$ revolution ($T_{\text{rot}} = T_{\text{rev}}$).
2. Assuming random arrivals, the expected value (average) is the midpoint of the uniform distribution: $\text{Average } T_{\text{rot}} = \frac{0 + T_{\text{rev}}}{2} = \mathbf{\frac{1}{2} T_{\text{rev}}}$

---

Question 2: Explain the difference between a "Track" and a "Cylinder" on a hard disk drive. Answer:

• Track: A single circular concentric recording ring on a single magnetic surface of a platter.
• Cylinder: The collection of all tracks across all surfaces and platters that share the exact same radial distance from the spindle center.
• Accessing data on different tracks within the same cylinder requires zero actuator arm movement (zero seek time), making cylinder-aligned file storage significantly faster.

Common Interview Traps

• The RPM to Milliseconds Conversion Trap: When calculating rotational latency from RPM: $\text{RPM} = 7200 \implies \text{rev/sec} = \frac{7200}{60} = 120 \implies 1\text{ rev} = \frac{1000}{120} \approx 8.33\text{ ms} \implies \text{Average Latency} = \mathbf{4.17\text{ ms}}$ Always check that you divided by $60$ and converted seconds to milliseconds.
• Overlooking Seek Time in Random Reads: When comparing HDD performance to SSDs, remember that SSDs have zero mechanical seek time ($0\text{ ms}$), while HDDs are throttled by $5\text{–}10\text{ ms}$ of physical arm inertia on every random read.

---

🗺️View Complete Operating Systems Roadmap & Syllabus

← Previous Topic8.5 Frame Allocation & ThrashingNext Topic →9.2 FCFS & SSTF Disk Scheduling