Input/Output Systems: Devices, Buffers, Scheduling & Performance

TAKEAWAYS:

• I/O devices are classified by speed (block vs. character), access method (sequential vs. random), and sharing (dedicated vs. shared).
• Device drivers act as translators between OS and hardware, handling interrupts and DMA for efficiency.
• Buffering (single, double, circular) reduces CPU idle time by staging data transfers.
• Scheduling algorithms (FCFS, SSTF, SCAN, LOOK) optimize disk arm movement—SCAN is best for high throughput, LOOK for fairness.
• Performance metrics: Throughput (I/O ops/second) and response time (latency) are critical for apps like Daraz’s order processing.
• Real-world tie: NTC’s fiber-optic backbone uses SCAN-like scheduling to route calls efficiently across exchanges.

1. I/O Hardware and Classification

I/O devices bridge the CPU and external systems. The OS manages them via device controllers and drivers.

Classification of I/O Devices

classDiagram
    class Device {
        <<abstract>>
        +speed: String
        +access: String
        +sharing: String
    }
    class BlockDevice {
        +speed: "High (e.g., SSD)"
        +access: "Random (e.g., disk)"
        +sharing: "Shared (e.g., HDD)"
    }
    class CharDevice {
        +speed: "Low (e.g., keyboard)"
        +access: "Sequential (e.g., printer)"
        +sharing: "Dedicated (e.g., serial port)"
    }
    Device <|-- BlockDevice
    Device <|-- CharDevice

Key Terms:

• Block devices: Transfer data in fixed-size blocks (e.g., disks, SSDs). Used by file systems.
• Character devices: Stream data (e.g., keyboards, printers). Used for real-time I/O.
• Shared vs. Dedicated: Printers are shared; GPUs may be dedicated.

Labelled diagram of a HDD showing platters, actuator arm, and read/write heads. (Image: Evan-Amos, CC BY-SA 3.0, via Wikimedia Commons)

2. Device Drivers and Interrupts

Device drivers are OS modules that control hardware. They handle:

1. Initialization: Configuring hardware on boot.
2. Command Execution: Sending I/O requests (e.g., read(2)).
3. Interrupt Handling: Responding to hardware signals (e.g., disk completion).

Interrupt-Driven I/O

When a device finishes an operation, it sends an interrupt to the CPU. The OS:

1. Saves the current process state.
2. Executes the Interrupt Service Routine (ISR) for the device.
3. Restores the process.

Example: eSewa’s payment gateway uses interrupts to log transactions in real-time.

sequenceDiagram
    participant CPU
    participant DeviceDriver
    participant Disk
    CPU->>DeviceDriver: Issue read(100)
    DeviceDriver->>Disk: Send I/O request
    Disk-->>DeviceDriver: Interrupt (I/O done)
    DeviceDriver->>CPU: Return data

3. Direct Memory Access (DMA)

DMA controllers transfer data without CPU intervention, reducing overhead.

• How it works:
  1. CPU sets up DMA controller with source/destination addresses.
  2. DMA transfers data directly to/from memory.
  3. Interrupts CPU only when done.

Real-World Use: Pathao’s ride-matching app uses DMA to stream GPS data from phones to servers without CPU bottlenecks.

4. Buffering Strategies

Buffering stages data to balance CPU and device speeds.

Types of Buffers

Example: Daraz’s order queue uses a circular buffer to handle spikes during sales.

stateDiagram-v2
    [*] --> BufferFull
    BufferFull --> BufferEmpty : Data read
    BufferEmpty --> BufferFull : Data written

5. I/O Scheduling Algorithms

Optimizes disk arm movement to minimize latency. Key metrics:

• Seek Time: Time to move the arm to a track.
• Rotational Latency: Time for the sector to rotate under the head.
• Transfer Time: Time to read/write the data.

Comparison Table

Worked Example: NTC’s call routing uses SCAN to minimize delays across exchanges.

• Scenario: Requests at tracks 10, 22, 20, 2, 40.
• SCAN Order: 10 → 20 → 22 → 40 → 2 (arm reverses at end).

6. Performance Metrics

7. Spooling

Simultaneous Peripheral Operations Online (SPOOLing):

• Uses a spool manager to queue print/job requests.
• Steps:
  1. Jobs written to disk (spool).
  2. CPU processes next job while device handles current one.
  3. Output written to disk until device is free.

Real-World Use: Banks use spooling to batch process loan applications overnight.

8. Error Handling in I/O

Common errors:

• Device Failure: Retry or notify user (e.g., printer out of paper).
• Data Corruption: Use checksums/CRC (e.g., WhatsApp’s message integrity).
• Timeouts: Requeue requests (e.g., Daraz’s failed payment retries).

Example: Ncell’s 4G network uses CRC to detect corrupted packets.

In the Real World

1. eSewa’s Payment Gateway:

   • Uses double buffering to stage transaction data before processing.
   • SCAN-like scheduling for database queries to minimize latency.

2. Daraz’s Order Fulfillment:

   • Circular buffers manage order queues during sales.
   • SSTF prioritizes nearby warehouses to reduce delivery time.

3. NTC’s Fiber-Optic Backbone:

   • SCAN algorithm routes calls across exchanges to balance load.
   • DMA streams voice data without CPU bottlenecks.

4. Khalti’s Transaction Logging:

   • Spooling batches transactions for audit trails.
   • Checksums verify payment data integrity.

Exam Tip

1. Diagrams are key: Draw disk arm movement (SCAN/C-SCAN) and buffer states in exams.
2. Compare algorithms: Memorize the pros/cons table for scheduling algorithms.
3. Real-world ties: Relate buffering to apps (e.g., YouTube’s double buffer) and scheduling to systems (e.g., NTC’s SCAN).
4. Formulas: Know throughput and response time calculations.
5. Deadlock hint: I/O deadlocks occur when processes wait for devices held by others (e.g., two printers sharing a spool).