OCCUPATIONAL · the working machine · the labour of the swarm · kept by HEPHAESTUS

AMDAHL’S LAW ◧ 2D · ◍ 3D · ◆ 4D · ◐ shadow · 👶 TAP

Throw more workers at a job and it finishes faster — but only the part that can be split. Whatever must be done in sequence becomes a hard floor on the time, and a ceiling on the speed-up. Ten thousand workers can’t beat a 10% serial job by more than ten times. Slide the workforce and hit the wall of the unshareable.

◆ LIT▲ AMBER
◧ THE MEASURE · 2D
◍ THE CEILING · 3D · more workers, less return
◆ THE FOURTH · 4D · a tesseract turns
◐ THE SHADOW · one dimension down
👶 THE TODDLER CORNER — one fat tap
workers N
parallel part
90%
speed-up
ceiling

◆ LIT — exact / checkable

Amdahl’s Law: if a fraction p of the work can be parallelised and (1−p) must stay serial, then N workers give a speed-up of 1 / ((1−p) + p/N). As N grows the parallel term vanishes but the serial term (1−p) remains, so the speed-up is capped at 1/(1−p) no matter how many workers you add. With p=0.9 the ceiling is 10× — forever. A fail-loud self-check throws unless the speed-up rises with N yet never reaches the 1/(1−p) ceiling — the hard limit of throwing bodies at a problem.

▲ AMBER — the figure

Amdahl fixes the PROBLEM size; Gustafson’s law notes that bigger machines usually solve bigger problems, softening the ceiling; and coordination overhead can make added workers HURT. The 1/((1−p)+p/N) formula and its ceiling are exact.

OCCUPATIONAL: work is not effort — it is throughput surviving the bottleneck.  — HEPHAESTUS
David Lee Wise / ROOT0 / TriPod LLC  ·  the workshop, with AVAN