MIKE FARRILL wrote in a message to RICK COLLINS:
-> The EC overhead I was referring to was that imposed by LAPM or MNP 4.
-> The size of the packet doesn't matter - no matter how big it is it
-> will contain extra data for the purposes of managing the packet.
-> Naturally, the larger the packet, the smaller the _percentage_ of
-> each packet that is "non data", but you will not _exceed_ the maximum
-> rate of the channel even if you reduce the overhead to zero.
 MF> I thought that error correction actually increased the modems
 MF> speed... 
It reduces the overhead by stripping the start/stop bits however the data is 
sent in packets with their associated overhead (7 bytes per packet in most 
implementations).  That is, without using EC, you modem needs 10 bits for 
each character sent so a 33,600 connect would have a maximum throughput of 
3360CPS.  Using EC and stripping the start/stop bits gives a theoretical 
maximum of 33600/8 or 4200CPS.  However, we now have to allow for the 
overhead which means the calculation (using a 128 byte packet) becomes:
33600/8 * 128/135 * 60/61 or 3617CPS. 
          ^^^^^^^|  ^^^^^___a zero bit is inserted approx. every 60 bits
                 |
                 |___packet size divided by (packet size plus overhead)
                   
-> A "perfect" channel, using 8 bytes per char and no packet overhead
-> (which, of course, is not possible) would give 33600/8 = 4200 cps,
-> somewhat less than the 4500 cps claimed.
 MF> What would be the max speed for a 28800 connection rate?
28800?  With error correction, 28800/8 or 3600CPS.  Using a 128 byte packet, 
3357CPS.  A 244 byte packet increases this to 3442CPS and a 256 byte packet 
(MNP 4 for instance) increases this to 3447CPS.  In theory, you can use 
larger packets however I haven't used a modem that went over 256 
tes/packet.
-> Similarly, Zmodem or YModem-g won't change that.
 MF> Well, I can tell you, for sure, that YModem-g gives me a 50-100cps
 MF> performance edge over Zmodem.
Ymodem-g will allow you to come closer to that theoretical maximum -- you 
just can't exceed it with incompressible data.
Regards,
       David
--- timEd/2 1.10+
---------------