MR> It's been suggested to me that possibly the Dynalink problems I'm having
MR> stems from it negotiating or training "too aggressively"
 MR> especially on poor
MR> phone lines.

 BG> Not so.  In fact, USRs (and presumably their clones)
 BG> actually connect low, but will shift upwards almost
 BG> immediately if the line will support it.

Hmmm, that doesn't get around the problem exhibited by some modems (notably
around here a large rack of Netcomm M34s) in timing out too fast in V.42
negotiation (thus obtaining non-EC connects, which I seem to remember was
Michael's problem, no?) because they accept a higher bit error rate as
worthy of granting (say) a 28800 connect than the calling modem does.

The solution for both my DPX596 and the Motorolas was to set them so that
they'd go for a lower speed (say one or two 2400bps steps) for a given line
quality, thus quickly negotiating a lower speed, but getting error
correction with these modems.  Apart from the link reliability, the overall
throughput is higher at say 26400, probably even 24000, with EC than at
28800 without.

It's possible that the Netcomms have a setting that will either be less
aggressive about it, or (more likely, being Rockwells) perhaps allow longer
for EC negotiation - but I've not been able to find out if that's so, as
yet.

MR> Does anyone know what S register setting will make a Dynalink train less
MR> aggressively.

 BG> There isn't one.  All you can do is try disabling the
 BG> higher symbol rates (say 3429 and 3200, for example) for a
 BG> lower connect speed.

That surprises me, on a modem of that supposed quality.  Ok, the Rockwells
don't appear to have any user-accessible BER adjustments either, but still
..

Perhaps Lewin can help with this, regarding how the Rockwells can be tuned
.. it's my theory (totally by smell) that Netcomm and maybe some other
modem chip packagers are not testing their modems nearly adequately on the
less than perfect lines that many of us enjoy, due to distance from
exchanges, water in the works, bad line connections, etc, with enough other
makes of modem, thus are setting their BER / line rate selection policy
toward only better lines.

If true, this is counter-productive.  Going sooner to a lower rate will
achieve higher connectivity with more modems on poor lines, without in the
least sacrificing connect rates on the best lines.  Again, 24000/Arq is a
hell of a lot better than 28800/None, let alone dropped connects.



Ian

--- MaltEd 1.0.b5