Using Training-Operator To Schedule Distributed Edge-Cloud Collaborative...
summary
Demonstrates integrating training-operator with KubeEdge’s Sedna to run distributed training across cloud and edge, extending Sedna beyond its original data-parallel federated learning to support model and pipeline parallelism and automated resource scaling. Implements a data-loader daemon plus init container that blocks workers until local datasets sync, uses the training-operator API for group scheduling and lifecycle operations, and performs gradient aggregation via a remote aggregator; claims reduced cross-region bandwidth and millisecond-level responses (1–10 ms) while enabling fault recovery.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:04Cloud native at the edge
- 2:57AI collaboration challenges
- 4:44KubeEdge architecture
- 8:40KubeEdge features
- 10:07Sida AI framework
- 11:34Training Operator with Sida
- 13:43Distributed training & federated learning
- 16:34Integration advantages
- 17:39Federated learning workflow
- 19:27Summary
related talks
32mFrom Ground to Space: Cloud-Native Edge Machine-Learning Case Studies with KubeEdge-Sedna
34mManaging Edge Nodes With “KubeEdge”
07mKubeEdge Everywhere: From Graduation To Global Adoption
31mIntro to KubeEdge: Kubernetes Native Edge Computing Framework
33mKubeEdge: From Fixed Location to Movable Edge, Latest Updates and...
25mManage Cloud Native LLM Workloads Across Edge and Cloud Seamlessly Using KubeEdge and WasmEdge
24mDemo of Kubernetes at the Edge for AI Model Deployments
27mManaging Large-Scale Edge Cluster Over Unstable Network with KubeEdge