BBR Bottleneck Bandwidth and RTT
Google's congestion control algorithm, which measures what the link can actually carry instead of waiting for packet loss to tell it. Released in 2016 and now everywhere.
Classic algorithms read a dropped packet as "the network is full", which was reasonable when buffers were small. With today's oversized buffers the queue can grow to hundreds of milliseconds before anything is dropped, so loss-based control fills those buffers as part of its normal operation. BBR instead builds a model of two quantities: the highest throughput observed and the lowest round-trip time observed. Their product gives how much data may be in flight, and the send rate is held right at the bottleneck's capacity without building a queue. The result is markedly higher throughput on long, lossy paths. Versions 2 and 3 have mostly been about not starving CUBIC flows.