A netkit paper measures the container networking penalty at 26 to 31 percent and recovers it in Cilium

Sources: write-up, paper, discussion
Summary: The paper measures container networking at 26 to 31 percent below host networking throughput and traces the gap to redundant per-CPU backlog queue traversals at veth pairs. Moving to netkit devices removes those traversals, and the recovery is reachable on Cilium v1.19.5 behind two Helm values. The write-up is dated 2026-09-22 and the paper it covers carries DOI 10.1145/3837779.3838164.
Why it matters: A throughput cost that has been treated as the price of container networking is traced to a specific queueing path and closed by two configuration values on a current Cilium release.