The current implementation of BBR replaces the "startup" code of BBR by Hystart when the latency is above a threshold. This was due to two perceived issues with the BBR startup:
it did not exit if detecting an increase in RTT, which could lead to high losses if the path buffer were small
it did not properly incorporate the link rate measurements, which leads to slow startup on long delay links.
Abandoning this PR. There does not seem to be a good way to make BBR startup performance match HyStart, probably because testing on a succession of epochs is too coarse compared to testing the RTT of every packet.
The current implementation of BBR replaces the "startup" code of BBR by Hystart when the latency is above a threshold. This was due to two perceived issues with the BBR startup: