This makes enormous sense in that a version of RED that does it only on
bytes barely budges when faced with pure ack packets rather than mixed
workloads. And certainly doesn’t account very well for asymmetric
regrettably, RED in linux is bytes, with an average packet size parameter. Twiddle that number as I may and I don’t get particularly good results. So writing/finding a version of red that comes closer to the ns2 model may help.