[e2e] Policing TCP flows

[Greg Minshall]

Dennis,

(Mark Handley pointed out to me that my math was faulty, and the 1/5th RTT 
implies 25 times the number of packets dropped, based on the "TCP equation".)

> I'll buy the factor, but I think there's something wrong that attributing
> great importance to small factors like that

i agree.  i'm not trying to bash policers.  i like policers.  if there's an 
argument about this point, you're having it with someone else.

> If you make minimizing congestion control packet loss rates your primary
> measure of goodness you'll end up concluding that infinite buffering and
> infintesimal MSSes are optimum, and that can't be right.

yes, absolutely.  i go so far as to say "i *like* dropping packets".

however, i think it is *natural* for people, in general, to consider that 
dropping a packet is "a bad thing".  your arguments of the absurdities help 
convince people otherwise.  as you go on to say, the goal is to keep the pipes 
filled, hopefully in a "fair" way.  the way we have chosen to do this, the way 
we think works and is cost effective, involves dropping packets (both as a 
signaling mechanism -- a role ECN may at least partially subsume -- and as a 
load shedding mechanism).  i'm not talking to Dennis now, but it is incumbent 
on us to make the case that there's nothing wrong with an honestly dropped 
packet.  (that's the "high order bit"; lower order bits include the work done 
up to the point the packet is dropped, trying not to let the pipe drain after 
the packet has been dropped, blah, blah, blah.)

> If you have
> both policers and shapers attempting to do this then over longer delay paths
> I think you'll find that both end up at a drop rate which differs by only
> a small factor, and which as a fraction of the packets being sent is
> quite modest in either case.  It is only on paths with a very short
> uncongested round trip that the policer loss rate starts to look stupid.

"small factor" can cover a multiple of multiples.  i think quite naturally 
(though incorrectly), people (sys admins, say) dislike dropping packets.  if 
they see even a factor of 2, they will complain ("you're *doubling* the bad 
metric!").  and, they may see a factor of 25 in a real world situation 
(assuming i've got the maths right).  i think for most people, 25 isn't a 
"small factor".  thus, i think one can't say "it's a small factor, forget 
about it".  i think (back on my soap box) one has to explain that dropping 
packets is neither inherently good nor inherently bad.

(some other time i'd love to discuss "everyone 'knows' what [a real queue] 
should behave like"...)

cheers, Greg