Community resourceWorksheet

OCR H446 2.2.2 Software pipelining

Part 11 of 14 · H446 2.2.2 · Computational methods

Software pipelining is routinely confused with processor instruction pipelining, and only one of them is what H446 2.2.2 is asking about here. Stage timings from a media service give students the arithmetic behind a bottleneck, and the worksheet keeps latency and throughput firmly separate throughout.

Students will:

  • explain stages, hand-offs and how different items occupy different stages at once
  • identify the bottleneck stage and estimate the best steady completion interval
  • separate the time taken by one item from the rate at which items complete
  • distinguish software pipelining from processor instruction pipelining
  • propose and justify a pipeline improvement, and name a correctness risk that overlapping introduces

Inside: 7 explanation cells, 1 multiple-choice question, 2 fill-in-the-blanks cells and 2 written answers. 20 marks, about 20 to 30 minutes.

Series: H446 2.2.2 · Computational methods, part 11 of 14.

Shared by Coding PathwayVerified teacher

  • 12 cells
  • About 30 minutes
  • CC BY-SA 4.0
  • Shared 31 Aug 2026
  • Updated 3 Sept 2026

Preview

The whole resource, exactly as a class sees it. Answers and marking are held back.

Software pipelining

A media service validates uploaded records, transforms them and produces summaries. A software pipeline passes the output of one process into the next and can overlap work on different items.

By the end, you will be able to

  • explain stages, hand-offs, filling and throughput;
  • distinguish latency for one item from throughput for many;
  • identify a bottleneck and propose a justified change;
  • distinguish software pipelining from CPU instruction pipelining.

Reactivate: a dependency identifies what must finish before another process can use its result.

Output from one stage becomes input to the next

Software pipeline timelinetime 1time 2time 3time 4time 5item A: validateA: transformA: summariseitem B: validateB: transformB: summariseitem C: validateC: transformC: summariseAfter filling, different items occupy different stages at the same time; the slowest stage limits throughput.

For one item, validate must finish before transform, and transform before summarise. Once item A leaves validate, that stage can start item B while A is transformed. After the pipeline fills, several items are being processed at different stages.

The latency of one item includes all its stages. Throughput is the rate at which completed items emerge. The slowest sustainable stage limits that rate because faster stages eventually wait or queues grow.

Worked timing model

Suppose stage times are validate 2 s, transform 5 s and summarise 3 s. One item needs about 10 s of processing. Once filled, the pipeline cannot complete items faster than approximately one every 5 s because transform is the bottleneck.

Speeding validate from 2 s to 1 s does not change the 5 s bottleneck. Splitting transform across suitable workers might improve throughput, but only if data can be distributed safely and ordering, communication and resource costs remain acceptable.

Multiple choice1 mark

Which change is most likely to improve steady pipeline throughput?

  • AOptimise a 2-second stage while a 5-second stage is unchanged
  • BAdd a decorative output field
  • CReduce the slowest sustainable stage time
  • DProcess only one item in the pipeline at a time
Fill in the blanks3 marks
A pipeline passes each stage’s gap 1 to the next stage, can overlap different gap 2, and has throughput limited by its gap 3 stage.
  • output
  • items
  • slowest
  • first

Guided calculation

A four-stage pipeline takes 1, 4, 2 and 3 seconds per item. Identify the bottleneck, estimate the best steady completion interval, then explain why the first result still takes roughly the sum of all four stages. Keep the terms throughput and latency separate.

Written answer5 marks

Analyse the four-stage pipeline described above.

State bottleneck, steady completion interval, first-item latency and one reason the estimate may be optimistic.

Students type their answer here.

Do not confuse two syllabus ideas

Here, software pipelining means the result from one software process feeds the next, with different data items potentially occupying different stages. Processor instruction pipelining overlaps fetch, decode and execute stages for machine instructions. Both overlap stages, but their level, data and controlling mechanism differ.

Written answer7 marks

A school analytics service has stages: import files (2 s), validate records (9 s), anonymise records (4 s) and calculate summaries (6 s). Propose and justify a pipeline improvement, then explain one correctness risk introduced by overlapping work.

Apply your answer to named stages and data hand-offs.

Students type their answer here.

Closed-book checkpoint

Complete each sentence from memory. There is no answer bank and correctness is held for teacher review.

Fill in the blanks4 marks
A software pipeline divides processing into connected checkpoint gap 1. Different data items may checkpoint gap 2 across them. The total time for one item is its checkpoint gap 3. The checkpoint gap 4 controls the best sustainable completion rate.

Review your understanding

Before submitting, check that you can explain the central distinction in your own words, expose the intermediate state that supports your answer and apply the method in an unfamiliar context.