1. Why edge computing? #
Addressing latency challenges in an ultra-connected world.
The end of full centralization #
Sending all the data from a self-driving car to a central cloud takes too long (latency). Edge computing moves computing power as close as possible to the data source: on the sensor or in a local mini data center.
2. Edge vs cloud: complementary #
The edge handles real-time decisions (immediate reaction). The centralized cloud processes aggregated data for training AI models or historical archiving. The two work in tandem.
3. IoT, 5G and Industry 4.0 #
The arrival of 5G boosts edge computing by enabling massive throughput. In smart factories, the edge can detect an anomaly on an assembly line in a few milliseconds, avoiding costly accidents.
Hands-on: deploying as close as possible to users #
Edge and latency.
Measuring the value of the edge #
# Mesurer la latence depuis un client
curl -o /dev/null -s -w "DNS: %{time_namelookup}s connexion: %{time_connect}s TLS: %{time_appconnect}s premier octet: %{time_starttransfer}s\n" https://app.exemple.fr
mtr -rwzc 20 app.exemple.fr
# Déployer une fonction à la périphérie (Cloudflare Workers)
npm create cloudflare@latest edge-demo && cd edge-demo && npx wrangler deploy
# Edge sur Kubernetes léger
curl -sfL https://get.k3s.io | sh - # k3s : cluster Kubernetes de quelques centaines de Mo pour l'edgeEdge-to-cloud solutions and modular data centers #
Industrializing the edge.
Offerings #
| Approach | Examples | For |
|---|---|---|
| Cloud extended on premises | AWS Outposts, Azure Stack Edge and Azure Local, Google Distributed Cloud | The same APIs as the cloud, as close as possible to the data |
| Lightweight Kubernetes at the edge | K3s, MicroK8s, KubeEdge | Stores, factories, remote sites |
| Modular data center | Prefabricated containers or racks (power, cooling, security built in) | Deploy capacity in weeks on an industrial site or in an isolated area |
An edge-to-cloud architecture processes and filters locally (latency, network outages, volume) and sends aggregated data, learning and centralized monitoring to the cloud.
Good to know: Monitoring an edge fleet relies on local collectors that buffer during outages (OpenTelemetry Collector, Alloy, Bindplane) and on remote fleet management.
