Technical explanation
C accurately describes eBPF host-routing. In a conventional datapath, packets may traverse substantial portions of the host networking stack and its iptables hooks even when Cilium performs routing decisions with eBPF. eBPF host-routing takes a more direct datapath, bypassing iptables and the upper host stack while providing a faster transition between the host and pod network namespaces. This reduces processing and context-switching overhead and can improve throughput and latency.
Option A describes load-balancing behavior rather than host routing. Backend distribution is implemented through Cilium’s service load-balancer maps and algorithms. Host routing can improve the path used by resulting packets, but it does not itself guarantee even backend selection.
Option B is the description of BIG TCP. BIG TCP increases the size of internal GSO and GRO packets to reduce stack traversal. Although BIG TCP requires eBPF host-routing in supported Cilium configurations, the two are distinct features.
Option D is overly specific and does not define the feature. Host routing optimizes compatible pod traffic generally, subject to kernel, kube-proxy-replacement, masquerading, netfilter, encryption, and integration constraints.
Official references
Cilium eBPF Host-Routing
Study Guide topic: eBPF host-routing, host-stack bypass, veth traversal, and performance.
