BBR

BBR stands for Bottleneck Bandwidth and Round-trip propagation time. It is a congestion-control algorithm for TCP developed by Google that seeks to improve network throughput while reducing latency. Unlike traditional loss-based approaches, BBR models the path between sender and receiver to estimate the available bottleneck bandwidth and the round-trip time, and then uses those estimates to pace and limit data transmission.

The algorithm operates by continuously estimating two key parameters: the bottleneck bandwidth and the minimum RTT along the path. Using these estimates, BBR adjusts its sending rate to match the network’s capacity while aiming to keep the queue at the bottleneck small. It cycles through phases that probe for more bandwidth and then drain queueing to measure current conditions, employing pacing to distribute data and avoid creating large buffers.

BBR is distinguished from conventional congestion-control methods such as CUBIC or Reno by relying on a model-based approach rather than primarily reacting to packet loss as a congestion signal. It strives to maximize throughput with low latency by preventing long queues, which can reduce buffering delays in many network environments.

BBR has undergone iterations, most notably BBRv1 and the later BBRv2. BBRv2 introduces refinements intended to improve fairness against loss-based traffic and to perform more reliably in wireless and variable-RTT networks. It has been implemented in the Linux kernel and deployed in various data-center and cloud environments, where it is used to optimize long-haul and intra-data-center connections.

In other contexts, BBR can be an acronym for terms or organizations outside networking; the intended meaning typically depends on the domain and usage.