🚀 go-pugleaf

In the literature on Microcom's "High Speed Error-Correcting Modems"
which are the ones which incorporate Class 6, it says the following:

"Additionally MNP Class 6 includes Statistical Duplexing(tm) which
allows the modems to allocate line bandwidth and data traffic to
achieve maximum throughput in both interactive use and file transfer.

                   ------------------------------------------\
                                                              \
Interactive                     9600 bps                      /
Communications    /------------------------------------------/
                 /
                 \              9600 bps
                  \------------------------------------------


                    _________________________________________
                                                             \
                                19,200 bps                    \
File                                                          /
Transfer            _________________________________________/
                   /               110 bps
                   \-----------------------------------------

Statistical Duplexing achieves full duplex performance using half
duplex modulation techniques. "

  It also says that the 19,200 bps performance throughput is reached
with 9600 bps transmission plus data compression, etc. (covered in my
previous posting) and the minimum packet size shown is 32 characters.
Note from the last lines in the quote above that the transmission is
half-duplex.

  My question is how any kind of packets, with half duplex, even with
"Fast Training" 9600 bps, can send interactive sessions (e.g., vi) and
give good performance?
  Back of the envelope calculation: 1 data character/packet + 4 other
characters/packet, with the packet sent at 9600 bps, and 1 millisecond
line turn around time gives 333 echoplexed characters/second.  And I
think that my assumptions may be too favorable-packet overhead is
likely to be >4 characters, and line turnaround is likely to be >1 ms.
  Maybe this speed is sufficient when a person is typing, and then 
dynamically switching to longer packets as soon as characters start 
flowing fast in either direction, e.g., for screen redraw, gives more 
than adequate performance.
  Anybody know?
--henry schaffer