[e2e] TCP flow count estimation

[Dennis Ferguson]

> if tcp is sending instead of MSS, alternate packets of size
> MSS-1, and size 1
> (well payload size)
> and every packet elicits an ack,
> it can use path char style estimation of the bottleneck line rate

I may not be understanding how pathchar works, but doesn't making
estimates this way require not only sending variable-sized packets
but also that they be sent slowly enough that their arrivals are
independent at the receiver in a statistical sense?  To put it another
way, if the path maintains order and you send a big packet followed
back-to-back by a small packet, the small packet can't possibly be
delayed less than the large packet (or they'd arrive out of order),
so samples like this aren't telling you anything.  Or if the packets
are sent spaced out, but insufficiently, you'll only see the small
packet's reduced store-and-forward delays up to the point where it
catches up to the big packet, after which the delays in the remaining
hops will have to increase to match the big packet if ordering is
maintained.

It hence seems that this would work reliably only if the packets are sent
really slowly, which may defeat the purpose one might have to do it in
the first place.  Certainly as a control signal its errors lead you in
the wrong direction.  If you send the packets too fast their delays will
become consistently more similar, which will encourage you to think you
can send the packets faster still.

Dennis Ferguson