No subject

which contribute to congestion, in the following sense: those packets
which arrive at the queue after the start of a busy period but before the
queue overflows. 

This suggestion comes from a quite different way of thinking about marking
to that of Floyd etc. It is based on economic ideas about fairness and
efficiency, and on the theory of effective bandwidths. (And it only makes
sense when queueing delay is less than RTT.) 

A more colourful way of putting it is this: If we only mark packets that
arrive after a packet has been dropped, we are closing the stable doors
after the horse has bolted -- and blaming the horses that we stable next. 
It would be facile of me to make such a remark without any theory
behind it; if anyone is interested, the theory is in
  "How to mark fairly"
  http://www.statslab.cam.ac.uk/~djw1005/Stats/Research/marking.html

Damon Wischik.