talk about latest rtt too

lib/quicly.c

@@ -3038,10 +3038,11 @@ static uint32_t calc_pacer_send_rate(quicly_conn_t *conn)
 {
     /* The multiplier uses a hard-coded value of 2x in both the slow start and the congestion avoidance phases. This differs from
      * Linux, which uses 1.25x for the latter. The rationale behind this choice is that 1.25x is not sufficiently aggressive
-     * immediately after a loss event. Following a loss event, the congestion window (CWND) is halved (i.e., beta), but the Smoothed
-     * Round Trip Time (SRTT) can remain high for a couple of round-trips since it is a moving average adjusted with each ACK
-     * received. Consequently, if the multiplier is set to 1.25x, the calculated send rate could drop to as low as 1.25 * 1/2 =
-     * 0.625. By using a 2x multiplier, the send rate is guaranteed to be no less than 1x. */
+     * immediately after a loss event. Following a loss event, the congestion window (CWND) is halved (i.e., beta), but the RTT
+     * remains high for one RTT and SRTT can remain high even loger, since it is a moving average adjusted with each ACK received.
+     * Consequently, if the multiplier is set to 1.25x, the calculated send rate could drop to as low as 1.25 * 1/2 = 0.625. By
+     * using a 2x multiplier, the send rate is guaranteed to become no less than that immediately before the loss event, which would
+     * have been the link throughput. */
     return quicly_pacer_calc_send_rate(2, conn->egress.cc.cwnd, conn->egress.loss.rtt.smoothed);
 }