Kubernetes networking is simple when all apps within a single cluster. But once applications span clusters and regions, connecting them often means adding layers of gateways, overlays, NAT, and routing control planes.

What if pod networking stayed simple: no per-cluster ingress or egress gateways, no host agent, and no overlay data path?

This talk shows how we built and deployed a global pod network using IPv6 and IPVLAN. Each pod receives a routable IPv6 address derived deterministically from its host’s IPv6 address, configured via CNI plugins without a node-level routing agent. The result is pod IPs that behave like real infrastructure addresses: reachable across clusters and regions, visible to service discovery, delivered through a high-performance IPVLAN data path.

We will cover the addressing model, why IPv6 makes it practical, how service discovery and routing works without relying on Kubernetes Services, kube-proxy or cluster DNS.