Intermediate

Edge Fleet Orchestration

Manage thousands of edge AI devices with Kubernetes at the edge, cloud IoT services, and purpose-built orchestration platforms.

Orchestration Platforms

⚙

K3s / KubeEdge

Lightweight Kubernetes distributions for edge. K3s runs in 512 MB RAM. KubeEdge extends cloud K8s clusters to edge nodes with offline autonomy.

☁

AWS IoT Greengrass

Run Lambda functions and ML models on edge devices with local messaging, OTA updates, and seamless AWS service integration.

💻

Azure IoT Edge

Deploy containerized workloads to edge devices managed from Azure IoT Hub. Native integration with Azure ML for model deployment.

K3s for Edge AI

Bash - K3s Edge Deployment
# Install K3s on edge device (Jetson, Raspberry Pi, etc.)
curl -sfL https://get.k3s.io | sh -s - --write-kubeconfig-mode 644

# Deploy AI inference pod with GPU access
kubectl apply -f - <<EOF
apiVersion: apps/v1
kind: Deployment
metadata:
  name: ai-inference
spec:
  replicas: 1
  selector:
    matchLabels:
      app: ai-inference
  template:
    metadata:
      labels:
        app: ai-inference
    spec:
      containers:
      - name: detector
        image: registry.local/detector:v2
        resources:
          limits:
            nvidia.com/gpu: 1
        ports:
        - containerPort: 8080
EOF

Fleet Management Considerations

AspectK3s/KubeEdgeIoT GreengrassAzure IoT Edge
DeploymentKubernetes manifestsComponent recipesDeployment manifests
Offline SupportFull (autonomous)Yes (local Lambda)Yes (local containers)
GPU SupportNVIDIA device pluginLimitedNVIDIA runtime
ScaleThousands of nodesMillions of devicesMillions of devices
Lock-inNone (open source)AWSAzure
✅
Best practice: Use K3s when you need GPU support, multi-container workloads, and vendor independence. Use cloud IoT platforms (Greengrass, IoT Edge) when you need to manage millions of simple devices and want deep integration with cloud AI services for model training and monitoring.