New BBS in Maryland

[Mouse]

> It's all about latency and jitter.  If you can keep the latency and
> jitter down... and consistent... modems will actually work pretty
> well over VoIP and you can sometimes pull off some fairly high data
> rates... If jitter on the link is very bad, good luck, even at 300
> baud.  Is this something usually made better by setting up QoS or
> traffic shaping?

"It depends."

At the really low data rates, like 110 or 300 bps (which =baud for
those rates), there's no negotiation and it's all about consistent
latency, which means low jitter.  (Consistent high latency is better
than inconsistent low latency, for this, so crank those jitterbuffer
sizes up - jitter is really just small fast fluctuations in latency,
and the line between jitter issues and latency issues is a fuzzy one.)

At higher rates, it matters more.  The usual 1200bps, for example, is
600 baud with two bits per baud and is like 110 and 300 in that there
is no negotiation; this means that high but constant latency doesn't
disturb it and turning the jitterbuffer up to what would normally be
ridiculous levels can fix a lot.  But it takes less to disturb 1200
than it does to disturb 300 or 110; bits are smaller, after all.  A
half-second of jitterbuffer delay is enough to pretty much wreck a
voice call but all it will do to a 300 or 1200 modem call is introduce
data latency (which may or may not matter, of course).

At the rates involving ECC and negoation between the ends, it really
depends on the codec.  Most traditional modem codecs are designed to
deal with the kinds of errors typically introduced by POTS.  Errors
typically introduced by VoI are very different and traditional modem
codecs generally fall flat on their metaphorical faces when faced with
them.  Recently, modem codecs designed to deal with VoI-style errors
have been designed, though I haven't seen them make it into the field
except for the special case of faxes.  I once did some experiments with
faxes using traditional codecs and I found that, as long as you have
absolutely zero packet loss and enough jitterbuffer to deal with
latency variations, they work great, but I found even a single lost
packet can take out several dozen lines of fax data.

I don't know how sensitive traditional modem codecs are to high (but
consistent) latency.  My tests didn't involve that.

So, if your QoS or traffic shaping operates to improve latency
consistency without generating ploss, it is likely to help.  If it
introduces ploss, it is likely to hurt.  (If it does neither, I find
myself wondering what it _is_ doing. :-)

/~\ The ASCII				  Mouse
\ / Ribbon Campaign
 X  Against HTML		mouse at rodents-montreal.org
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B