Adom PCL Node Data Profile

Install?

The physical shape of the data produced by the Adom Industries NSF PCL Test Bed node: every instrument that generates data, the file types each writes, and the rate it writes them at.

adom-wiki pkg install adom/pcl-data-profile

Latest: v0.1.5, published

Contents

README

markdown

Data Generation at the Adom PCL Node

Adom Industries, NSF PCL Test Bed node, Fort Worth, Texas.

The instruments at our node that generate data, the file types each produces, and the rate at which they produce it today and projected as we scale.

Our node is an electronics-prototyping cloud lab. We plan on about 1.2 TB per workcell per day: about 200 GB of video and about 1 TB of sensor data. Video is steady; sensor data is bursty and is the highest value per byte.

We run 20 workcells today, producing about 24 TB/day. We expect 100 by the end of 2027 (about 120 TB/day, or 44 PB a year) and project 1,000 by 2029.


Instruments and file types

Per-workcell instruments

Instrument File types Rate
High-resolution live cameras (multiple 4K60 streams per workcell) .mkv, .ts (H.264/AVC) records 4K (3840×2160) at 60 fps, ~50–80 Mbps; simultaneously streams 1080p or 1440p at 60 fps to users
Microphone .wav 44.1 kHz, 176.4 kB/s
Power sensor .wav audio-rate waveform stream
BME690 — barometric pressure, temperature, humidity .csv, .txt, .bin, .json, .jsonl low rate, typically 1–10 Hz
BMI270 — 6-DOF IMU (accelerometer + gyroscope) .csv, .txt, .bin, .json, .jsonl up to 1.6 kHz ODR

The camera stream is a live view for remote users and is not retained; the 4K60 record is the archived artifact.

Bench instruments

Instrument File types Rate
Oscilloscopes — MXR series .csv, .wav, .png 1 Gbps streaming
Spectrum analyzers custom binary (.bin); exports .csv, .sta 5 Gbps streaming
FLIR thermal cameras .jpg (radiometric) frame rate set per experiment
µV/µA voltage and current sensing .csv, .bin, .json, .jsonl, .wav sample-rate dependent

These are our highest-rate instruments by interface speed. They capture in short bursts around a device under test rather than streaming continuously. Sensor data, led by oscilloscope captures, is most of our daily volume: about 1 TB of the 1.2 TB per workcell.

Custom user sensors

Our node exists to let other labs prototype electronics, so a user can put any component on a board we fabricate and read it back through our stack. The lists above are our fixed inventory; the sensors a user brings are open-ended. Their formats follow the same menu: .csv, .bin, .json, .jsonl, .wav, and images.


Rates

Per-unit, per-day estimates for each instrument. Assumes 8 active hours per day.

Estimated data volume per instrument, GB/day per unit on a log scale

Instrument Per unit, per active day
4K60 camera ~216 GB (~650 GB if run continuously)
Power sensor ~17 GB
Microphone ~5 GB
BMI270 IMU ~46 MB at 100 Hz, ~740 MB at 1.6 kHz (binary; ~10× as JSONL)
BME690 under 30 MB
Oscilloscope (MXR) burst-limited; single-GB to tens of GB on a light day, up to ~1 TB
Spectrum analyzer burst-limited; tens of GB, TB-scale if streamed
FLIR thermal MB for single shots, up to ~40 GB continuous
µV/µA sensing sub-GB logging to ~70 GB at 100 kHz
Custom user sensors user-defined; unbounded

We plan per workcell, not per instrument: ~200 GB/day of video and ~1 TB/day of sensor data. The ~1 TB is a planning figure driven by bursty oscilloscope captures, so it sits above what the light-day values in this table add up to.


Current vs. projected

Raw volume tracks workcell count.

Today End of 2027 Projected by 2029 Beyond 2029
Workcells 20 100 1,000 10,000
Video (~200 GB per workcell-day) 4 TB/day 20 TB/day 200 TB/day 2 PB/day
Sensor data (~1 TB per workcell-day) 20 TB/day 100 TB/day 1 PB/day 10 PB/day
Total ~24 TB/day ~120 TB/day ~1.2 PB/day ~12 PB/day
Per year ~8.8 PB ~44 PB ~438 PB ~4.4 EB
Sustained average ~2.2 Gbps ~11 Gbps ~111 Gbps ~1,111 Gbps

28 Gbps is our potential uplink once the factory is fully built out, not what is installed today.

A 28 Gbps uplink is ~302 TB/day if saturated, which covers the raw capture of roughly 250 workcells. Raw capture from 1,000 workcells would need about four times that, and from 10,000 about forty times.

So raw capture does not all go out. It lands in local caching at our Cloud Lab, and we publish derived metadata. We expect 50:1 to 100:1 compression from raw data to derived metadata, which would put 1,000 workcells at about 12 to 24 TB/day, well inside the uplink. The question is "what do we keep, what do we reduce at the edge, and what do we ship."


On these numbers

The instrument inventory, the file types, and the native interface rates are firm. The daily volumes are estimates: they assume 8 active hours per day, and the 50:1 to 100:1 compression is an expectation we have yet to measure. Actual volume depends on real duty cycles, which we will have once InstaPCB goes live in early 2027. This page is versioned and will be updated in place with measured figures then.


Contact: Kyle Bergstedt (co-PI, AI lead) — [email protected] · John Lauer (PI) — [email protected]