Edge-Fog-Cloud Latency Trade-offs
Compare response time, WAN traffic, model quality, and tier fit for one IoT workload.
Edge, fog, and cloud are placement trade-offs, not a speed ladder.
Pick a workload, then compare how response time, WAN traffic, and model quality proxy change when the same decision is made at the device edge, a nearby fog node, or a cloud service.
Animated Tier Comparison
Step through capture, local decision, fog hop, cloud hop, and return. The cards below update with the same formula used by the visual path.
The teaching model estimates request-response time. It includes capture, network transfer, queueing, and processing; it does not model every protocol handshake or radio retry.
Response-Time Stack
How The Model Works
Each tier uses the same structure:
response_ms = capture + transfer + propagation + queue + processing
Transfer time uses bytes and link throughput:
transfer_ms = bytes x 8 / (Mbps x 1,000,000) x 1000
The page uses 1 KB = 1024 bytes and 1 MB = 1024 KB. Network and processing values are teaching estimates; a production design should replace them with measured traces.
Tier Selection Checklist
| Tier | Use When | Main Risk |
|---|---|---|
| Edge | Control deadlines are tight, privacy matters, or WAN connectivity is unreliable. | Limited compute and smaller local model context. |
| Fog | Several devices need local coordination or aggregation inside one site. | Gateway capacity and local network congestion. |
| Cloud | Batch analytics, fleet-wide models, deep storage, or loose response budgets dominate. | WAN latency, cloud traffic, data exposure, and recurring cost. |
Practice Checks
- Choose Safety loop, then move the response budget from 20 ms to 100 ms. Explain whether the recommendation changes and why.
- Choose Camera triage and increase payload size. Watch which tier becomes limited by transfer delay before processing delay.
- Choose Fleet analytics and lower the quality target. Decide whether a fog or edge summary becomes acceptable.
Technical Notes And Next Steps
- The quality proxy is a relative teaching score for model capacity and available context. It is not measured accuracy, precision, recall, or validation loss.
- Queue delay is deterministic here so learners can compare settings. Real systems have jitter, retries, congestion control, and scheduler effects.
- Use this workbench with Edge-Fog-Cloud Latency Simulator, Architecture Comparison, and Edge vs Cloud Decision Tool.