Week 4 Bench: Timing a comparison

Measuring how long a string comparison takes, and watching the measurement get more honest as the sample size grows

15 March 2027

Before the Bench#

Have a timing harness ready in any language: something that runs a comparison function many times and records elapsed time per call, not just total time.

In the Bench#

Time two versions of the same secret-comparison function — one that returns as soon as it finds a mismatched byte, one that always checks every byte — against inputs that agree with the secret for a controlled, varying number of leading bytes. Plot mean time against agreement length. The early-exit version should show a trend; the constant-time one should not. Noise will be large relative to the effect on a shared machine, which is itself the lesson: single measurements lie, and the statistics are not decoration.

Afterwards#

The Trace Analysis Lab is due this week. It asks you to do the same kind of reading — trace to mechanism to conclusion — on material you have not seen annotated in advance.