Edge computing processes data near the place where it is produced or used, such as on a device or a local gateway, rather than sending every task to a distant data centre.
Decide which work belongs near the source
A device can filter readings, detect an event or run a model locally, while a central service stores selected records or manages updates. The useful boundary depends on response time, connectivity, available computing resources and the amount of data that needs to move.
NIST’s edge AI programme identifies constraints including resources, communication, privacy and security. Moving computation changes these trade-offs rather than removing them.
Example: count items on a production line
A local camera system could count items and send totals to a reporting service. If raw images are unnecessary, the design can avoid transmitting them. Logs, sampled images and remote support access may still expose information, so check the complete data flow.
Define what happens when connectivity fails: whether counting continues, how long results are stored and how missing or repeated uploads are reconciled.
Operate the devices as a fleet
Track device identity, software versions, storage limits and update failures. A locally fast decision is useful only if the device is configured correctly and its inputs are valid.
Cloud infrastructure may support central management and reporting. Edge computing describes where selected work executes; it does not require abandoning cloud services.