a deep understanding of bbr
bbr should be applied at the situation that the bottleneck buffer size is much smaller the BDP. In this case, BBR can achieve 200% higher goodput than cubic. While in the case of deep buffers, cubic can improve goodput by 30% compared to bbr.
BBR’s packet loss can be several orders of magnitude higher than that of Cubic when buffers are shallow. It is because BBR’s configuration maintains 2*BDP of data in flight.
when the loss rate exceeds a point, which is at around 20%, the goodput decreases significantly. The point is dictated by the maximum pacing_gain when probing for more bandwidth.
The reason why BBR maintains 2*BDP packets in flight is that one BDP is budgeted for the network capacity, and the other is to deal with delayed/aggregated ACKs.