Kubernetes Networking Infrastructure Offload
Talkat KubeCon + CloudNativeCon North America 2022 · on Friday, 2022-10-28· at 16:00· 330 AB, Huntington Place: 1 Washington Blvd·
summary
Panel argues moving Kubernetes networking data plane into infrastructure accelerators yields concrete gains: use of IPU/DPU and SR‑IOV switchdev offloads policy, routing and load balancing from host CPUs; Calico adds a P4-based Calico data plane and IPDK provides agent/manager pieces; measured lab numbers show OVS on host uses 32 cores and 8 Gbps while DPU offload uses 5 cores and sustains 50 Gbps vs 8 Gbps with ~4× lower latency and stronger isolation at cluster scale (15,000 nodes × 200 pods).
AI-written summary · about
chapters (8)
- 0:00Start
- 1:21What is Infrastructure?
- 2:51Why Separate?
- 5:35What is Offload - Networking Speak
- 7:13Objectives
- 8:35Goals/Requirements
- 11:02Potential High Level Target Architecture
- 31:35Summary
related talks
24mUsing Kubernetes with Data Processing Units to Offload Infrastructure
30mAll You Can Eat Networking. Kubernetes Goodness for the Hungriest Workloads
26mEnhancing K8s Networking with SmartNICs
40mKubernetes Networking Panel: Collaborate on the Spec, Compete on Execution
40mNoverlay Networks for Health and Wellbeing
27mMake Kubernetes Networking Ready for World Class AI and HPC ...
26mRun your Kubernetes cluster on OpenStack in Production
35mPanel: Kubernetes Networking: Present and Future