Learn edge processing vocabulary: edge vs. fog vs. cloud computing, edge inference, latency-sensitive workloads, CDN edge workers (Cloudflare Workers, Lambda@Edge), and edge Kubernetes (K3s).
0 / 5 completed
1 / 5
The distinction between edge computing and fog computing is:
Edge: computation happens on the device itself or a directly attached gateway (e.g., an ML inference chip on a camera). Fog (Cisco's term): a local network tier (e.g., a factory-floor server) that collects data from many edge devices, applies further processing, and reduces the volume of data sent to the cloud. Both reduce latency and bandwidth compared to cloud-only architectures.
2 / 5
Edge inference refers to:
Edge inference is enabled by model compression techniques (quantisation, pruning, knowledge distillation) and dedicated hardware (NPUs, TPUs, Arm Ethos, NVIDIA Jetson). Use cases: real-time object detection in security cameras, wake-word detection in smart speakers, anomaly detection in industrial sensors — all requiring sub-100ms responses that cloud round trips cannot reliably provide.
3 / 5
Cloudflare Workers and AWS Lambda@Edge are 'CDN edge workers'. What characterises them?
Cloudflare Workers run in V8 isolates (not containers) at 300+ PoPs globally, starting in microseconds. Lambda@Edge runs at CloudFront PoPs. Use cases: JWT validation at the edge (no origin hit for auth), geolocation-based redirects, request rewriting, image optimisation, and serving personalised content from edge KV stores — all reducing latency for global users.
4 / 5
K3s is described as a 'lightweight Kubernetes distribution for the edge'. It differs from standard Kubernetes by:
Standard Kubernetes has significant resource overhead (etcd, multiple control plane pods). K3s combines kube-apiserver, kube-scheduler, kube-controller-manager, and containerd into one binary and uses SQLite (or embedded etcd) for state — running on as little as 512 MB RAM. Rancher's K3s enables workloads like retail kiosks, in-vehicle systems, and industrial controllers to be managed with familiar Kubernetes tooling.
5 / 5
A 'latency-sensitive workload' is best suited for edge processing rather than cloud because:
Human reaction time is ~150ms; industrial PLC control loops operate at 1-10ms; autonomous vehicle decision cycles require <10ms. Cloud round trips (even at 30ms) are too slow and variable for these cases. Edge processing eliminates the WAN hop — decisions are made locally, with the cloud used only for non-time-critical tasks like model updates or historical analytics.