Level-3 Distillation Cost Calculation Audit

Level-3 Distillation Cost Calculation Audit

Ada re-derives this chapter’s own numbers step by step, at full precision

foundations
math-foundations
calculation-audit
analytics-ml
beginner
Ada ADA · CALCULATION AUDIT

Level-3 Distillation Cost Calculation Audit

The chapter streams a 100-sensor fleet to the cloud at about $721/month, then applies Level-3 edge processing — a 512-point FFT sending only the top 20 frequency bins — to cut the bill to about $28/month, a 96% reduction. That distillation keeps exactly the peak frequencies a bearing fault lives in. This audit works the Level-3 distillation bill line by line.

Companion to the chapter Edge in the IoT Reference Model — every number here comes from that chapter.

— a 100-sensor vibration fleet, raw vs FFT peaks, ~4 minutes

Level-3 distillation is a claim about bytes. Here is the fleet bill in strict decimal units (GB = 109 bytes), so the 96% cut is reproducible line by line.

See the relationship before changing it

The figure reads from left to right. The blue card is edge feature rate. The middle card applies this page's rule. The green card is monthly volume. Walk the arrows once: set the input, apply the rule, then read the result with its unit.

The retained audit below checks several chapter fixtures. This model keeps those stated values fixed and changes only edge feature rate, so the numeric fixture does not switch without explanation.

Edge feature rate changes monthly volume An input card leads through the rule month = bytes/s x 2,592,000 s / 1,000,000,000 to the monthly volume result. INPUT PAGE INPUT APPLY THE RULE predict calculate check units OUTPUT RESULT
Walk the arrows. Sending more retained FFT features raises the cloud bill in direct proportion.

Derive the baseline in four named moves

  1. 1

    Name the input. The chapter baseline is 78125 B/s.

  2. 2

    Name the relationship. month = bytes/s x 2,592,000 s / 1,000,000,000

  3. 3

    Substitute with units. 78,125 x 2,592,000 / 1,000,000,000 = 202.5 GB

  4. 4

    Read the result. Keep the unit beside the value. Use it only inside the technical boundary on this page.

Predict, then change edge feature rate

Try Predict the direction of month = bytes/s x 2,592,000 s / 1,000,000,000. Test another edge feature rate, then compare monthly volume.

78125 B/s
Chapter baseline
Monthly volume

Observe Sending more retained FFT features raises the cloud bill in direct proportion. Reset edge feature rate to 78125 and compare monthly volume.

Explain Sending more retained FFT features raises the cloud bill in direct proportion.

Check yourself

What should you do before trusting a moved-control result?
Answer: Predict its direction, apply the shown relationship, keep the units, and reset to the worked baseline.
What does this small model leave out?
Answer: Only edge feature rate moves here. Field effects named in the technical boundary stay fixed.

Honesty note. This teaching model changes only edge feature rate. It is not a field prediction.

TryThe chapter streams a 100 -sensor fleet to the cloud at about $721/month , then applies Level-3 edge processing — a 512-point FFT sending only the top 20 frequency bins — to cut the bill to about $28/month , a 96% reduction. Use Check derivation.
ObserveThe displayed ledger resolves 100, $721/month, 512-point FFT, 20, $28/month at full precision. This audit works the Level-3 distillation bill line by line. Check derivation shows this.
Explainthe spectral summary keeps exactly the peak frequencies a bearing fault lives in while cutting the monthly bill from about $778 to $30 — distillation is worth 96% here precisely because the diagnostic signal is a handful of bins, not the full 10 kHz waveform. Check derivation confirms it.

1. Raw is samples × bytes × sensors × seconds.

Per sensor 10,000 samples/s × 2 B = 20,000 B/s; fleet × 100 = 2,000,000 B/s = 2.0 MB/s.

month = 2,000,000 B/s × 2,592,000 s = 5.184 × 1012 B = 5,184 GB = 5.18 TB

At $0.15/GB that is $777.60/month. (This audit works in strict decimal GB throughout; the chapter uses binary 1024-based KB/MB/GB and its own precise fleet rate of 1,950 KB/s = 1.9 MB/s, landing on $721 — both are internally consistent, they simply define "GB" differently.)

2. The FFT window sets the send rate.

512 samples ÷ 10,000 Hz = 0.0512 s = 51.2 ms, so 10,000 ÷ 512 = 19.53 FFTs/s. Top-20 bins as 40 bytes: 40 × 19.53 = 781.25 B/s per sensor.

3. Scale the edge stream back to a bill.

×100 = 78,125 B/s → month = 202.5 GB → $30.38. Reduction = (5184 − 202.5) ÷ 5184 = 96.1% (a 25.6× cut).

Stage Arithmetic Monthly Cost
Raw fleet 2.0 MB/s × 2,592,000 s 5,184 GB $777.60
FFT edge fleet 78,125 B/s × 2,592,000 s 202.5 GB $30.38
Reduction (5184 − 202.5) ÷ 5184 96.1% 25.6×

What the audit buys you: the spectral summary keeps exactly the peak frequencies a bearing fault lives in while cutting the monthly bill from about $778 to $30 — distillation is worth 96% here precisely because the diagnostic signal is a handful of bins, not the full 10 kHz waveform.

Technical boundaries
The distillation bill deliberately does not simulate FFT compute energy, overlapping windows, spectral leakage, missed fault features, price tiers, or retransmissions; it prices the stated raw samples and top-20-bin payloads only.

Work the audit first, then check the displayed derivation.

Every number above is taken from the chapter’s own examples and re-derived step by step.