What the 10x Deadband Number Really Depends On
Ada re-derives this chapter’s own numbers step by step, at full precision
ADA · CALCULATION AUDIT
What the 10x Deadband Number Really Depends On
The chapter reports a 10x reduction from edge deadbanding: 100 sensors at 200 bytes each produce 20 KB/s, and suppressing 90% leaves 2 KB/s. But the reduction factor is 1 / (1 - s), so a ten-point swing in the suppressed fraction moves the headline from 5x to 100x. This audit reproduces the number and shows what the 10x deadband number really depends on.
Companion to the chapter Lab: Edge-Fog Computing — every number here comes from that chapter.
Ada: The section reports a “10x reduction” from edge deadbanding and then, to its credit, insists the number “is only trustworthy if it survives scrutiny.” Let me reproduce it and then show exactly how fragile that headline is.
The raw stream first:
100 sensors x 1 msg/s x 200 bytes = 20,000 bytes/s = 20 KB/s- Suppress 90%, so 10% survives:
20,000 x 0.10 = 2,000 bytes/s = 2 KB/s - Reduction:
20,000 / 2,000 = 10x
The arithmetic holds. But the reduction factor is 1 / (1 - s) for a suppressed fraction s, and that curve bends sharply:
s = 0.80gives1 / 0.20 = 5xs = 0.90gives1 / 0.10 = 10xs = 0.95gives1 / 0.05 = 20xs = 0.99gives1 / 0.01 = 100x
A ten-point swing in the suppression fraction moves the headline from 5x to 100x. And this is only the mean: if a burst suppresses nothing, the peak upstream rate snaps back to the full 20 KB/s regardless of the 10x average — and the peak is the capacity a link must actually be provisioned for.
The design meaning is why the chapter demands stress and tail measurement rather than a quiet-minute average: the deadband ratio is hypersensitive to the one assumption nobody measures — how often the signal truly holds still — so an honest lab reports the suppression fraction it observed under load, not a single round multiplier.
Every number above is taken from the chapter’s own material and re-derived step by step.