Implementation of pacing. When the connection is sending at a rate lower than permitted by the congestion controller, pacer is responsible for spreading out the outgoing packets across the entire RTT. Technically the pacer provides two functions: - computes the number of packets that can be sent now - computes the time when another packet can be sent When a new flow starts, or when the flow is not pacer-limited, the pacer limits the window to: max(INITIAL_WINDOW, pacingRate / timerFreq) timerFreq is the best timer resolution we can get from the selector. pacingRate is N * congestionWindow / smoothedRTT where N = 2 when in slow start, N = 1.25 otherwise. After that, the window refills at pacingRate, up to two timer periods or 4 packets, whichever is higher. The time when another packet can be sent is computed as the time when the window will allow at least 2 packets. All methods are externally synchronized in congestion controller. Ideas taken from: https://www.rfc-editor.org/rfc/rfc9002.html#name-pacing https://www.ietf.org/archive/id/draft-welzl-iccrg-pacing-03.html