Why Two Copies Outperform One: Designing bpf_sock_splice_pair for High‑Speed TCP Loopback
The article examines the design of the new BPF function bpf_sock_splice_pair for intra‑host TCP communication, explains why a single‑copy implementation is suboptimal, introduces a ring‑buffer based two‑copy approach with optional busy‑polling, and presents benchmark results showing up to 7× throughput gains over the baseline.
