Zero-overhead Container Networking with eBPF and Netkit
summary
Benchmarks show container TCP throughput rises from ~66% to about 90 Gbit of a 100 Gbit link by replacing veth with a new eBPF-driven device type; the kernel adds Netkit devices plus tcx and helpers like BPF redirect peer and BPF redirect neighbor to perform host routing in eBPF and avoid twice-through-stack handling and per-CPU backlog queues. The implementation matches host-network latency and throughput and is available on kernels 6.6+ and via psyllium.
AI-written summary · about
chapters (7)
- 0:00Introduction and context
- 1:36Container networking overhead
- 5:09Bypassing host routing
- 7:10Addressing queuing delays
- 8:35TCX and kernel improvements
- 10:25Netkit for container speed
- 12:00Performance results and conclusion
related talks
31mUnleashing the Tides of Kubernetes Networking by Removing kube-proxy
37mEfficient and Flexible Virtual Machine Networking Through eBPF
37mEfficient and Flexible Virtual Machine Networking Through eBPF
46mConquering the Linux network stack: decapsulating packets with eBPF at 6Mpps+
35mAn Introduction to eBPF - Superpowers for Networking, Observability & Security
23mAn Intro to eBPF with Go : The Foundation of Modern Kubernetes Networking
25meBPF Strengthens SR-IOV to Be Powerful
35mLiberating Kubernetes From Kube-proxy and Iptables