Edge Aggregation
Summarising or filtering telemetry at the device or gateway so that only what is needed leaves, which is usually the difference between a viable and an unaffordable design.
A sensor sampling at one hertz produces eighty-six thousand readings a day. A thousand sensors produce eighty-six million. Transmitting and storing all of it, when the overwhelming majority is unchanged from the previous reading, is a cost with almost no corresponding value.
Aggregation reduces at the source. Send a summary per interval — minimum, maximum, mean, count — rather than every sample. Send only on change beyond a threshold. Detect the events of interest locally and send those, with a short window of raw data around each. Buffer and batch to amortise transmission overhead.
The judgement that has to be made deliberately is what is discarded forever. Raw data not transmitted cannot be recovered, and a future question that needs it cannot be answered. The usual compromise is a rolling local buffer of raw data that can be requested on demand when an anomaly is being investigated, with aggregates sent continuously.
The related benefit is autonomy: a device that computes its own thresholds continues to act correctly while disconnected, which a design that ships everything to the cloud for evaluation does not.