SB> Can anyone tell me whether it is neccessary to enable
 SB> retraining at both ends of a link to actually enable it, or
 SB> just one end?

It depends on the modems and chipsets/firmware used, some use quite
different algorithms for line quality versus channel bitrate adjustments. 
Some are more or less adjustable by the user, and some are not.

In Rockwell-chip modems %E0 both disables local and rejects remote
retrains, as I understand it.  Noone seems to use that, except perhaps on
leased lines.  %E1 accepts and responds to both fallback and fallforward
requests from the remote modem, but only initiates fallback requests when
detecting unacceptable signal quality (at a desired bit error rate), while
%E2 allows the modem to both accept and initiate fallforward and fallback
requests as necessary.

Other modems allow user setting of acceptable bit error rate threshold,
action on poor signal quality, the period of detecting improved signal
quality before renegotiation attemps, etc.  Differing revisions of some
modems/chipsets may also behave quite differently with different types of
line impairments.

 SB> OR, is it either the calling end, or the receiving end whos
 SB> retrain on/off setting decides whether retraining is
 SB> enabled for that link?

I can see why test results might surprise.  Many people (understandably,
given the usual poor standard of technical documentation in modem manuals)
seem to think that %E1 will stop their modem falling forward when requested
to do so by a remote (%E2-using, in Rockwell parlance) modem, and perhaps
avoid using it due to that misunderstanding.  It's demonstrably not so.

 SB> All the modem books I've seen dont say much in the way of
 SB> specifics about this.

Indeed, had me baffled for ages.  Remote Rockwell-chip modem (older Maestro
144FM, as it happens) was then on very poor, quite variable lines.  Left to
its own devices with %E2 set, it would merrily spend most of its time
requesting and getting rate changes (much faster than full retrains),
moving next to no data meanwhile.  Many failed connects, much cash being
wasted.

Using %E1 on the 'problem' modem allowed my modem to control rate
adjustment attempts, keeping line speed down and successful throughput
rates up, while allowing fallforward when, as often happened, the line was
seen to improve later on in the call, allowing shifting up to 12k or better
from as low as 4800 sometimes.  The line concerned is now very good, but
not all are ..

Keeping a (local, at least) connection healthy matters much more than top
speed, at least when you're interested in minimising cost.  Other people
out on poorer lines have found %E1 at their end greatly improving
connectivity, even with some assorted more recent Rockwell gadgets, to here
anyway ..

Cheers, Ian

--- MaltEd 1.0.b5