[e2e] Policing TCP flows

[Greg Minshall]

the TCP connection gets the same amount of throughput ("goodput") in both 
cases, but the "badput" is radically different.  policing incurs lots more 
badput.

if you look at plots from traces, what you see is that with policing, the RTT 
is very low, so the "probe-until-drop" cycle is very short, so you get lots of 
these cycles in the same elapsed time.  with shaping, the RTT is 
(artificially) high, so the "probe-until-drop" cycle is very long and you get 
fewer of these cycles in the same elapsed time.  since the amount of data 
transmitted is the same (in this "same elapsed time"), with policing you drop 
lots more packets.

if you hate dropped packets, you'll hate policing.

(this is all without ECN; i'm not sure how it would change with ECN, maybe for 
the better.  this might be the killer app for ECN!)

cheers, Greg Minshall