Unleashing the Tides of Kubernetes Networking by Removing kube-proxy
summary
Replacing kube-proxy with Cilium’s eBPF data path claims to restore near–bare‑metal throughput: a 100 Gbps baseline drops to ~63 Gbps with default Cilium, enabling kube-proxy replacement and BPF host routing raises throughput, and adding netkit removes namespace context‑switching to hit wire speed. It moves service LB, NAT and connection tracking into kernel maps, but watch socket load‑balancing compatibility and long‑lived TCP sockets (dead backends require socket termination or disabling socket LB); works on bare metal and managed clusters like EKS.
AI-written summary · about
chapters (13)
- 0:00Introduction and roadmap
- 1:12What is kube-proxy
- 2:39Why eBPF matters
- 4:30Default Cilium installation
- 8:07Enabling proxy replacement
- 11:38Socket load balancing issues
- 14:09BPF masquerade feature
- 16:42Host routing in eBPF
- 19:12Performance benchmarks
- 20:36Netkit innovation
- 22:24Future default settings
- 23:50Recap and resources
- 24:12Q&A session
related talks
35mLiberating Kubernetes From Kube-proxy and Iptables
29mSimplifying the Networking and Secur...
14mZero-overhead Container Networking with eBPF and Netkit
25mThat’s Just My Cup of Tea: Configuring Cilium for Performance and Scale
26mSupercharge Kubernetes Networking with Cilium and IPv6
07mLightning Talk: Taming Noisy Network Neigbours with Cilium and eBPF
35mNext-level Kubernetes Networking with Cilium - Nico Meisenzahl and Philip Welz
22mRole of eBPF in Kubernetes Networking and Security