Panel: Kubernetes Networking: Present and Future
summary
Panel summarizes concrete progress: DRA GA for GPUs in production enables scheduling of NICs, GPUs and FPGAs with NUMA/PCI affinity and bandwidth hints; Multus remains the de facto multi‑NIC solution while a network‑interface‑first API and a multi‑network subproject aim to replace it; EndpointSlice zone‑local services GA 1.30–1.31 provides topology‑aware traffic to cut inter‑zone egress costs; several migration and resource‑claim templating gaps remain and ongoing work targets conformance and a practical migration path.
AI-written summary · about
chapters (13)
- 0:00Welcome to panel
- 1:02Current networking APIs
- 3:52Dynamic networking resources
- 7:23Telco and AI networking overlap
- 10:14Common APIs for portability
- 13:06Q&A on Kubernetes networking
- 14:10Topology-aware routing
- 16:34Secondary interfaces
- 21:28Multus to DRA migration
- 26:03Custom DRA drivers
- 28:31Protocols beyond TCP/SCTP
- 31:17Secondary networking framework
- 33:23Multi-dimensional networking challenges
related talks
32mPanel: Are We There Yet? Native Kubernetes Networking for Telecom Workloads
27mMake Kubernetes Networking Ready for World Class AI and HPC ...
37mPanel Discussion: Cloud Native Networking State of the Union
40mKubernetes Networking Panel: Collaborate on the Spec, Compete on Execution
19mSIG-network: Updates and Future Directions
31mBreaking the Single-Network Barrier: A Cross-Domain Look at…
30mKubernetes WG Device Management - GPUs, TPUs, NICs and More With DRA
11mSIG-NETWORK: Updates and Directions