FR: BBRv3

[passcod]

There seems to be a stalled or abandoned effort upstream, we could look into that.

[passcod]

After some research:

• v2 looks very bad for performance, very "fair" but also completely dies whenever there's loss, which makes it, frankly, useless
• v3 is better than v2 but worse than v1 on throughput, and tries to be fairer