Making On-Prem Bare-Metal Kubernetes Network Stack Telco Ready
summary
Deutsche Telekom runs Kubernetes nodes as first‑class participants in the IP EVPN fabric so hosts peer BGP EVPN and learn tenant and cluster VXLANs directly, turning pod and node IPs into routable EVPN type‑5 host routes. Implementation uses FRR plus Linux VRFs, a new netplan/netplaner for VRF/VXLAN wiring, and a Kubernetes network-operator to push FRR config; result: no overlay mesh, simpler route automation at hundreds‑to‑thousands‑cluster scale, SR‑IOV remains unsolved so DPUs and Cilium/AF_XDP are being evaluated and the code is open source.
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:42Organization
- 4:11Das SCHIFF
- 5:32Cloud-Native in a Telco World
- 6:00Mission
- 6:52Problem Statement
- 7:40Introducing IP Fabrics
- 8:47Traffic Flow in an IP Fabric
- 9:24Layer 2 View
- 9:56Making the Host part of our IP Fabric
- 10:54VXLAN View
- 11:54Solution
- 12:46Network Connectivity from the Host
- 14:43Network Operatorii
- 15:46Netplanner
- 17:47Caveats
- 19:04Conclusion
related talks
09mLightning Talk: Hacking Kubernetes Networking to Create Telco Solutions
38mVPCs in Kubernetes
28mScalable and Multitenant Networking in XDP and Kubernetes Operators
34mBridging Baremetal and Virtualized Clouds: Operating Telco-Grade Kubernetes at Scale
26mAdvanced Network Capability for NFV on Kubernetes
35mMigrating the Next Generation Mobile Core Towards 5G with Kubernetes
26mRun your Kubernetes cluster on OpenStack in Production
20mKubernetes at the Edge for Telcos