Previously, congestion control algoritm selection was done via a simple method of looking for packet loss on a per-connection basis and applying BBR if a limit was surpassed.
This approach has been replaced by using a reinforcement learning based algorithm which seeks to choose the congestion control algorithm that approaches optimal performance for the link; minimum RTT and delivery rate tuned to fill the pipe. See bpftune-tcp-conn(8) for details.
The behaviour under loss is similar to before - BBR is selected - but it is selected based upon its optimal behaviour rather than via a hand-coded algorithm. In normal network conditions, different algorithms can be selected; choices available currently are
Previously, congestion control algoritm selection was done via a simple method of looking for packet loss on a per-connection basis and applying BBR if a limit was surpassed.
This approach has been replaced by using a reinforcement learning based algorithm which seeks to choose the congestion control algorithm that approaches optimal performance for the link; minimum RTT and delivery rate tuned to fill the pipe. See bpftune-tcp-conn(8) for details.
The behaviour under loss is similar to before - BBR is selected - but it is selected based upon its optimal behaviour rather than via a hand-coded algorithm. In normal network conditions, different algorithms can be selected; choices available currently are