Adapting Kubernetes to Constrained IP Address Environments
Talkat KubeCon + CloudNativeCon North America 2019 · on Tuesday, 2019-11-19· at 14:25· Room 16AB - San Diego Convention Center, Mezzanine Level·
summary
Kubernetes treats every pod as a first‑class IP so clusters quickly exhaust space: giving each node a /24 means a 100‑node cluster needs roughly a /17 and more than 85% of clusters run >16 pods/node. Practical fixes include tuning kubelet maxPods and node IPAM to shrink per‑node CIDRs (yielding 50–99% IP savings), using island mode to reuse cluster CIDRs across clusters, and changing masquerade logic to rely on node‑CIDR detection so fragmented or discontinuous ranges can be added dynamically.
AI-written summary · about
chapters (12)
- 0:00Intro
- 2:50Kubernetes Networking Model
- 4:48Implications of Ks Networking Model
- 6:39Node & Cluster. The Networking Abstractions...
- 8:06Two Deployment Modes
- 9:47Customer and Deployment perspective
- 11:48Solution: Optimize IP Utilization
- 15:20Solution: Dynamic & Discontiguous Pod CIDRO
- 18:35Solution: Clusters As Islands
- 21:53Solution: Clusters As Hybrid Islands
- 25:10Kubernetes Improvements
- 28:32IPv6 - Food for thought
related talks
22mDiscontiguous CIDRs for Dynamic Cluster Scaling
42mKubernetes Networking at Scale
26mIPv6 First, Not Just Ready: Kubernetes Without IPv4 Using Cilium
06mLightning Talk: Optimizing SNAT Port and IP Address Management in Ku...
18mThe Road to IPv6 Support in...
15mDemo: Resource(Cost) optimization on planet scale Kubernetes clusters.
38mDynamic Pod Resource Boundary Adjustment in Web Scale Clusters
38mDynamic Pod Resource Boundary Adjustment in Web Scale Clusters