100Gbit/S Clusters With Cilium: Building Tomorrows Networking
Talkat KubeCon + CloudNativeCon North America 2022 · on Thursday, 2022-10-27· at 16:30· 252 AB, Huntington Place: 1 Washington Blvd·
summary
Cilium implements an IPv6-only data plane using eBPF at XDP, adds net4↔6 gateways (stateful and stateless), and a DNS64 flow for legacy IPv4 services; kernel 5.19’s Big TCP support is enabled in Cilium 1.13 to lift the 64k MTU limit. A new meta device replaces veth to push BPF into the device and avoid kernel queuing. Lab results show iperf bulk throughput rising from 52→60 Gbit/s, P99 latency dropping 2.2×, and request‑response transactions improving ~32%.
AI-written summary · about
chapters (15)
- 0:00100Gbit/s clusters with Cilium
- 1:02IPv6 origins with Cilium
- 2:51Why users adopt IPv6 in Kubernetes
- 4:59IPv4 to IPv6 in modern clusters
- 8:10IPv4 to IPv6 gateway options
- 10:37IPv6-only clusters and IPv4
- 12:24Demo: IPv4/IPv6 gateways
- 13:48Introducing Big TCP
- 17:57Big TCP efficiency
- 19:42Big TCP performance gains
- 20:45High-scale data plane with Cilium
- 21:49eBPF host routing
- 22:55Meta device for pod networking
- 25:04Cilium’s high-scale components
- 25:45Closing & Q&A
related talks
26mIPv6 First, Not Just Ready: Kubernetes Without IPv4 Using Cilium
36mCilium: Welcome, Vision and Updates
25mThat’s Just My Cup of Tea: Configuring Cilium for Performance and Scale
04mClosing Remarks
07mWelcome + Opening Remarks
28mA Decade of Cilium Around the World
29mThe Swiss Army Knife of Cloud Native Networking
24mHigh-Scale Networking for ML Workloads With Cilium