Which kube-proxy mode is selected by default, and how does it affect latency?

The goal is to determine which kube-proxy mode is active when both iptables and IPVS are available, and how that choice influences network latency and throughput in large clusters.

Uncertainty stems from inconsistent default behavior across distributions and Kubernetes versions, with no explicit documentation stating the selection rule, and performance gains reported up to 30% with IPVS but varying with topology.

Which mode does kube-proxy select by default in a given environment? What is the impact of that selection on latency for clusters with more than 10,000 pods? How does kernel version support affect the reliability of IPVS versus iptables?