measure the latency in the queue (from ingress to egress, via time stamping on entry and checking the timestamp on exit
when latency exceeds tartget, think about dropping a packet
after latency exceeds target, drop a packet at the HEAD of the queue (not the tail!)
if that does not fix it, after a shorter interval (inverse sqrt), drop the net packet sooner, again at the HEAD
keep decreasing the interval between drops until the latency in the queue drops below target
we start with 100ms at the interval of the estimate and 5ms as the target (for 4Mbit/s and up); 10Gb/s needs a smaller target.
below 4Mbit/s ... well, we don't know