Week 6 Lab: Lab 6 — Starving the Long Job

Build a scheduler that starves someone, watch it happen, then fix it and measure exactly what the fix cost you.

29 March 2027

Before the lab#

Bring a laptop. Any language. If you would rather do it on paper with cards, that works and several people prefer it.

In the lab#

Part one: make it fail. Implement strict shortest-job-first on a stream where short jobs keep arriving and one long job is waiting. Run it long enough that the long job’s wait becomes indefensible. Screenshot it. This is the failure mode the lecture asserted and it is much more vivid when it is your own scheduler doing it.

Part two: fix it. Add ageing — a job’s effective priority improves the longer it has waited. You choose the function. Linear in waiting time is the obvious starting point; some of you will try something sharper.

Part three: price the fix. This is the actual work. Measure, for strict SJF and for your aged version:

• mean wait
• 95th percentile wait
• the single worst wait in the run

Then tune your ageing parameter and plot mean against worst-case as you sweep it. The curve you get is the exchange rate between the average and the tail for your particular system, and it is the single most useful artefact you will produce this semester. Every scheduling argument you have for the rest of the course is a question about where on that curve to sit.

There is no correct point on it. There is only a choice, and the curve makes the choice visible instead of implicit.

Afterwards#

Submit nothing from the lab itself. The ordering-rule assignment is due Friday at 17:00, and if your proposal involves any kind of prioritisation, the curve you plotted today is the evidence that you understood what you were proposing.