Device, edge, fog and cloud: where IoT data is processed
In short: IoT data can be processed on the device, at the edge (a nearby gateway), in the fog (site-level servers) or in the cloud. Work that needs fast responses or must keep running offline belongs close to the device; heavy analytics and long-term storage suit the cloud. Most systems split work across several tiers.
Four places to compute
Not every reading needs to travel to a distant data centre. Moving computing closer to devices can cut delay, save bandwidth and keep sensitive data on site. But local hardware is harder to manage and protect. Knowing the four tiers lets you place each job sensibly.
| Tier | Where | Good for | Limits |
|---|---|---|---|
| Device | On the sensor or controller | Thresholds, filtering, safety cut-offs | Tiny CPU and memory |
| Edge | Gateway or small computer near the devices | Protocol translation, buffering, local rules, video analytics | Physically exposed, needs patching |
| Fog | Servers at a site or region linking many edges | Site-wide coordination, aggregation | More infrastructure to run |
| Cloud | Provider data centre | Long-term storage, fleet dashboards, model training | Latency, internet dependence, data transfer cost |
The terms overlap in practice. "Fog" was popularised to describe a layer of distributed computing between edge and cloud; many vendors now simply say "edge" for anything outside the central cloud. In exams, define your terms before you use them.
Deciding where a job runs
- Latency: must the response happen within milliseconds (stop a conveyor)? Run it on the device or edge.
- Connectivity: must it work when the internet is down? Keep it local.
- Bandwidth: is the raw data large, like video? Process at the edge and send only results.
- Privacy: should raw data stay on site? Summarise locally.
- Scale and history: do you need months of data from thousands of sites? Use the cloud.
A worked example
A warehouse has 40 cameras and 300 temperature sensors. Each camera feed is analysed on an edge box that counts pallets and detects blocked fire exits, sending only events to the cloud instead of continuous video. Temperature sensors report to a gateway that raises a local alarm if a freezer warms up, even during an internet outage, and uploads five-minute averages. The cloud holds a year of history and compares all the company's warehouses. Each tier does what it is best at.
Security at each tier
Every tier you add is another place to secure. Edge boxes sit in plant rooms and car parks where they can be stolen or tampered with, so use full-disk encryption, secure boot, locked enclosures and remote management. Authenticate every link between tiers with certificates, not shared passwords, and send logs from edge and fog nodes to a central collector so they are not blind spots. Design for the edge to fail safe if it loses contact with the cloud.
Tip: a simple rule of thumb is "decide locally, learn centrally": safety and fast decisions near the device, fleet-wide analysis in the cloud.
Practice
Why might video analytics run at the edge rather than in the cloud?
Which tier is best for training a machine learning model on a year of data from 50 sites?
Name one security challenge that is greater at the edge than in the cloud.
Key takeaways
- Processing tiers: device, edge, fog and cloud.
- Put fast, safety-critical and offline-tolerant work close to devices.
- Use the cloud for long-term storage, fleet analytics and model training.
- Edge processing saves bandwidth and can protect privacy.
- Every tier needs authentication, logging and physical protection.
Original content written by Mohamed Fayas for FM TecHub. © 2026 FM TecHub. You're welcome to share a link to this page; please don't copy the text elsewhere without permission.