TCP pacing
Spreading packets evenly across the round-trip time instead of firing them in bursts as soon as the window allows. A prerequisite for modern congestion control to work at all.
A classic TCP sender is fundamentally jerky: acknowledgements arrive clumped, the window opens, and a whole batch of segments goes out at line rate. Along the path, a single bottleneck with a shallow buffer is enough for such a burst to drop packets entirely needlessly, even though the average rate was far below capacity. Pacing instead computes an inter-packet interval from the window divided by the estimated RTT. In Linux this is expressed as sk_pacing_rate and enforced either by the fq queue or by the stack itself, and network cards have begun offering offload for it. For BBR it is not an optimisation but a precondition, since the whole algorithm is built on sending at a computed rate rather than filling a window.