Edge computing runs code near where data is produced or used, on the device, a gateway or a small local site, instead of in a distant data centre. It cuts latency, saves bandwidth, keeps working when the link is down and can keep raw data local, at the price of managing many small, unreliable places.
Key concepts
- Placement: deciding what runs on the device, at the edge and in the central cloud, by latency, bandwidth, cost and privacy.
- Latency budgets: the speed of light and network hops set a floor that only proximity lowers.
- Disconnected operation: work continues offline and synchronises later, so conflicts must be handled.
- Constrained resources: limited power, memory and cooling favour small models, compact runtimes and efficient code.
- Fleet operations: deploying, updating and monitoring software across thousands of sites you cannot visit.
Vocabulary
- Edge node: a computer at or near the data source.
- Gateway: a device linking local devices to a larger network.
- CDN: a content delivery network, caching content in many locations near users.
- Inference at the edge: running a trained model locally rather than sending data away.
- Store and forward: keeping data locally until the link is available.
What to study, in order
- Networking basics: latency, bandwidth and what a round trip costs.
- Run a small service on a single-board computer and call it from another machine.
- Add local processing, such as filtering sensor readings, and send only the summary upstream.
- Make it survive a dropped link: queue, retry, synchronise and resolve conflicts.
- Practise remote deployment of a container to several devices, with a rollback path.
Related tags
#iot #cloud #containers #distributed-systems #embedded-linux