Breaking the Single-Network Barrier: A Cross-Domain Look at…
summary
Kubernetes multi-networking is shown as solved in part but fragmented: a production video AI pipeline at Expo 2025 uses RDMA over secondary network for GPU-direct transfers, many deployments use Multus + NADs but APIs differ per distribution, and routing, DNS, health checks and IPAM still break. The DRA work lets schedulers pick device-level attributes instead of nodes, and WG Device Management adds resource claims, consumable bandwidth and MAC/IP reporting, but a common network object or class is still being designed.
AI-written summary · about
chapters (7)
- 0:00Multi-network challenges
- 2:07Real-world multi-networking
- 6:25Multi-networking in Kubernetes
- 11:03Multi-networking limitations
- 14:17API standardization for multi-networking
- 17:51Community contributions
- 19:03Q&A
related talks
32mPanel: Are We There Yet? Native Kubernetes Networking for Telecom Workloads
35mPanel: Kubernetes Networking: Present and Future
46mMulti - Networking Kubernetes Containers with CNI
27mMake Kubernetes Networking Ready for World Class AI and HPC ...
09mMulti-Cluster Networking: A Technology-Agnostic Network Operator for K8 Federation
35mMultiple Networks for Kubernetes Workloads
37mPanel Discussion: Cloud Native Networking State of the Union
28mScalable and Multitenant Networking in XDP and Kubernetes Operators