I/O techniques: programmed I/O, interrupts, DMA, buffering and spooling

0 · Recall Warm-up

Answer from memory first. Answers are in the Answer Vault at the bottom.

1. SSTF total for the running queue? from Lesson 6.2.2
2. Average rotational latency at 6000 rpm? from Lesson 6.2.1
3. What is an interrupt? from Lesson 0.1.2

1 · The Hook

A kettle can be watched until it boils (wasting your time), or it can whistle when ready (you do other things meanwhile). Better still, ask a helper to pour all the cups and tell you when they're done. Those are programmed I/O, interrupt-driven I/O and DMA. GATE asks what fraction of the CPU each one uses.

3 · Core Concept

Predict First — the answer

4 µs out of every 100 µs = 4%: 25 times better.

In plain words

A device controller has a few registers (data, status, command). The CPU can keep checking the status register (polling), let the device interrupt when it's ready, or let a DMA controller move a whole block into memory by itself and interrupt only once at the end.

1. Programmed I/O (polling)

The CPU loops reading the status register until the device is ready, then moves each byte itself. Simple, but the CPU is busy waiting for the whole transfer.

2. Interrupt-driven I/O

The CPU starts the operation and does other work; the device interrupts when it has data (or finished). The CPU pays an interrupt overhead per transfer unit. CPU fraction used = (transfers per second) × (overhead per transfer). Example: 2,000,000 bytes/s moved as 4-byte words → 500,000 interrupts/s; at 1 µs each → 50% of the CPU.

3. Direct memory access (DMA)

• The CPU tells the DMA controller: source, destination, byte count. The DMA controller moves the whole block between device and memory and interrupts the CPU once per block.
• While copying, the DMA controller uses the memory bus, sometimes delaying the CPU's own memory accesses: cycle stealing.
• CPU cost per block = setup + one interrupt. Example: 2,000,000 B/s in 4,000-byte blocks → 500 blocks/s; at 20 µs each → 1% of the CPU.
• Cycle-stealing fraction = (words per second) × (memory cycle time). 500,000 words/s × 100 ns = 5%.

4. Buffering, caching, spooling

• Buffer: memory holding data in transit (e.g. speed mismatch between a modem and a disk). Double buffering: fill one buffer while the other is being processed.
• Cache: a fast copy of data kept for reuse (not the only copy).
• Spooling: a queue of complete jobs for a device that can't interleave, like a printer: each print job is saved to disk and printed in turn.

5. Blocking, non-blocking and asynchronous calls

• Blocking: the process waits until the I/O completes (simplest; the process goes to the Waiting state).
• Non-blocking: returns at once with whatever is available.
• Asynchronous: returns at once; the process is told later when the full transfer finishes.

4 · Visual

Who moves the data, and how often is the CPU disturbed?

Programmed I/O : CPU polls + moves every byte        CPU busy 100% of the transfer
Interrupt I/O  : device interrupts per byte/word      CPU pays overhead per unit
DMA            : DMA controller moves the block       CPU pays setup + 1 interrupt per block
                 (steals memory cycles meanwhile)

5 · Worked Examples

Example 1 Easy

A device produces 10,000 bytes per second, byte by byte. Interrupt overhead is 4 µs. Fraction of CPU time used in interrupt mode?

Solution

10,000 × 4 µs = 40,000 µs per second = 4%.

Example 2 Medium

In the Easy example, how many times better is interrupt mode than programmed (polling) mode, where the CPU is busy for the whole transfer?

Solution

100% / 4% = 25 times.

Example 3 GATE

A DMA controller transfers 4-byte words at 2,000,000 bytes/s, stealing one 100 ns memory cycle per word. What percentage of memory cycles does the CPU lose?

Solution

500,000 words/s × 100 ns = 0.05 s per second = 5%.

6 · GATE Traps

• Interrupt cost scales with the number of transfers, not bytes, if each interrupt moves a word or block.
• DMA interrupts once per block, not per byte.
• Spooling ≠ buffering: spooling queues whole jobs for a non-sharable device.
• Units: 1 µs = 10⁻⁶ s; 1 ns = 10⁻⁹ s.

7 · Memory Hook

"Poll = watch the kettle; interrupt = it whistles; DMA = a helper pours and tells you once."

8 · Connect the Dots

Phase 6 is complete, and with it the whole syllabus. Phase 7 puts everything together: the master rule sheet, mixed practice, the GATE pattern drill and the final test.

9 · Self-Test

GATE format. Try every question before opening the vault.

Q1 · NAT · 1 mark(s) — Interrupt mode

A device produces 20,000 bytes per second, byte by byte; interrupt overhead 5 µs. Percentage of CPU time used?

Numerical answer: ________

Q2 · NAT · 2 mark(s) — Gain

A device delivers 10,000 bytes per second (one at a time). Programmed I/O keeps the CPU busy for the whole transfer; interrupt overhead is 4 µs per byte. How many times better is interrupt mode?

Numerical answer: ________

Q3 · NAT · 2 mark(s) — DMA blocks

A device transfers 2,000,000 bytes per second using DMA blocks of 4,000 bytes; each block costs the CPU 20 µs (setup and interrupt). Percentage of CPU time used?

Numerical answer: ________

Q4 · MCQ · 1 mark(s) — Spooling

Several processes print at the same time, and each document comes out whole, one after another. This is achieved by:

1. (A) polling
2. (B) caching
3. (C) double buffering
4. (D) spooling

Q5 · MCQ · 1 mark(s) — DMA

With DMA, the CPU is interrupted:

1. (A) once when the whole block has been transferred
2. (B) never
3. (C) for every word
4. (D) for every byte

10 · Rule Card

• Programmed I/O: CPU polls and moves data (busy waiting).
• Interrupt I/O: CPU fraction = transfers/s × overhead per transfer.
• DMA: one interrupt per block; cycle stealing fraction = words/s × memory cycle.
• Buffer (in transit), double buffering, cache (copy for reuse), spool (queue of jobs for a non-sharable device).
• Blocking, non-blocking, asynchronous I/O calls.

11 · Concept Ledger

Unlocked in this lesson. From here on, these terms may be used.

device controllerstatus/data/command registersprogrammed I/O (polling)interrupt-driven I/Ointerrupt overheaddirect memory accesscycle stealingbufferingdouble bufferingcachingspoolingblocking, non-blocking and asynchronous I/O