Kubernetes Federation for Edge
summary
Describes a production design from NVIDIA that manages thousands of intermittent edge Kubernetes nodes by using a pull based federation where a federation controller publishes CRs and H controllers on edge nodes pull and materialize them; uses reverse SSH tunnels with an SSH server pod behind a load balancer to expose logs and services across NATs; layers a service mesh with iptables init containers to route app traffic and enable federated learning; demonstrated with a deep learning app but operational edge HA and hands‑off backup remain work in progress.
AI-written summary · about
chapters (10)
- 0:00Intro
- 0:42Agenda
- 1:38Edge Scenarios
- 2:37Al workload is a "bit" different
- 3:30Al Deployment@Edge
- 5:03Federation Architecture (Edge)
- 7:33Expose Services - Edge to World
- 9:08Feature Missing
- 9:44Service Mesh From Apps
- 10:29Is Edge Compute Still Primitive
related talks
11mKubernetes Federation for Edge
34mManaging Edge Nodes With “KubeEdge”
27mBuilding an Edge Computing Platform for Network Services Using Cloud Native Technology [I]
19mKeynote: Kubernetes Federation
46mCluster Federation in Kubernetes: Past, Present and the Future
36mEdge Computing in Kubernetes: Best Practices for the Future
32mKubeEdge DeepDive: Extending Kubernetes To the Edge With Real-World Industry Use Case
32mEdge-Native: The New Paradigm For Operating And Developing Edge Apps