← Back to the chapter · Previous detail

Technical detail · Applied: Edge Workload Placement

Timing limits and evidence lost during reduction

A cloud round trip is typically tens to hundreds of milliseconds; local edge processing can be sub-millisecond to a few milliseconds. That gap forces a millisecond-scale safety cutoff to run locally. It cannot wait for fleet analysis, Level 4 persistence, or operator confirmation.

Bandwidth is both a limit and a cost. Raw high-rate streams are often infeasible and expensive to transmit. Filtering, aggregating, and extracting features let useful compact results travel; there is no universal cloud rule rejecting every message above one kilobyte. Gateways can retain and replay data, so reduction does not require deleting everything locally.

The discarded information must be unnecessary for the decision being defended. One-minute averages may suit billing or environmental trends while hiding brief bearing-fault spikes, spectra, or phase relationships needed for diagnostics. Review which features matter before deleting source windows. Retain spikes or spectrum evidence, or raw windows for replay; small cloud storage alone does not prove the reduction preserved fault evidence.

Keeping raw voice, video, or health data local reduces disclosure through transmission. Sending only event clips or derived results can protect privacy, but incident review still needs timing, thresholds, confidence, and local clock evidence. Document what each reduction loses and when richer evidence is retained temporarily. Edge Stream Window Contracts supplies the window semantics and bounded state behind those summaries.

← Back to the chapter · Previous detail