Path: utzoo!news-server.csri.toronto.edu!rutgers!mit-eddie!bbn.com
From: pplaceway@bbn.com (Paul Placeway)
Newsgroups: comp.dcom.modems
Subject: Re: Testing actual throughtputs
Message-ID: <63156@bbn.BBN.COM>
Date: 10 Mar 91 04:35:35 GMT
References: <5051@mindlink.UUCP>
Sender: news@bbn.com
Lines: 36

Harvey_Taylor@mindlink.UUCP (Harvey Taylor) writes:

<     <curmudgeon mode on>
<     It really bugs me to see companies advertising 57.6K bps throughput
<  when they know damn well that the data types which allow 4:1 compression
<  are so rare, as to make their ad misleading.
<     <curmudgeon mode off>

Microcom is guilty of this (MNP5 won't do anything like 2:1 average),
but you really can get 4:1 *average* compression out of the newer
techniques (both LZ-ish and statistical).  (The best of statistical or
dictionary (LZ) methods are around 5:1 for USENet batches these days.)

As a brain-dead test a while ago, I ran script(1), and then did about
2 hours compile/edit/test work over the phone, ended the script, and
did "compress -v typescript".  Guess what: 3.9:1 compression average.

V.42bis uses a better method than 16-bit LZW, so it should do better
still.  Also note that this is with an ANSI-style terminal (Macintosh
Kermit :-)), and the ANSI sequences are kinda verbose (although they
make up for it by being parametric), so your mileage may vary, but
probably not by much.

Of course to actually see 4:1 average compression, you have to be able
to get better than 4:1 throughput on the "good" sequences, which means
a serial connection at better than 4 times the underlying connection
speed or big buffers in the modem and 4 times the underlying speed (4
* 14.4k == 57.6 kbps for v.32bis, more like 115.2 kbps or better for
V.FAST).

Don't confuse MNP5 hype with v.42bis hype.  The former is generally
untrue, but the latter isn't far off.

			-- Paul Placeway
	If you thought decoding a 512-point constellation is hard, try
	decoding human speech!