TCP CUBIC
Linux default congestion control since 2.6.19 (2006). Loss-based: increases cwnd via cubic function, halves on packet loss. Good on high-BDP links (long-distance). Standard in Linux, macOS, Windows.
Improvement over Reno: faster recovery after loss, less aggressive at high RTT. Inflicts buffer bloat on "dumb" buffered links (cable modems) — runs until packet loss → fills entire buffer → latency spikes. BBR (Google 2016) tried to replace with rate-based approach. Still highest deployment volume; CUBIC is "the default that just works". Tunable: tcp_cubic.beta, but rarely touched. cat /sys/module/tcp_cubic/parameters/.