Reimagine the Multi-Cluster
summary
A multi-cluster federation implemented with Lico runs on Kubernetes and uses virtual nodes to present peripheral clusters as schedulable nodes. A demo shows a central server and edge clients on Aruba Cloud Kubernetes where clients run Flower federated training locally for three rounds, aggregate model weights centrally over an automatic VPN, with no application changes, encrypted authenticated traffic, and privacy by design by moving computation to the data.
AI-written summary · about
chapters (9)
- 0:00Multicluster Kubernetes intro
- 3:34Lico project for federation
- 4:38Building multicluster infra
- 5:42Cluster peering setup
- 7:05Pod distribution on virtual nodes
- 9:57Argo CD app deployment
- 10:59Federated training with Flower
- 12:47Privacy and flexibility benefits
- 13:09Q&A
related talks
39mMulti-clusters made easy with Liqo: getting rid of your clusters by keeping them all
36mKubernetes Multi-Cluster Made Easy - Federica Ciuffo and Marco Ballerini
42mSIG Multi-Cluster Deep Dive
04mProject Lightning Talk: Empowering Federated Learning with Multi-Cluster Management for Privacy and Efficiency
19mKeynote: Kubernetes Federation
31mFederated Machine Learning with Kubernetes
18mAbstracting Multi-Cluster Topologies With CAPI and Linkerd for Internal Developer Platforms
39mMulti-Cluster Ops in a Hybrid World