[asterisk-dev] New chan_iax media-only transfers: questions

[Stephen Davies]

On 11/05/06, Kevin P. Fleming <kpfleming at digium.com> wrote:
> When I mentioned this to Mark while he was proposing the idea, he
> basically said he didn't see any reason to continue sending _any_ full
> voice frames, that the receiver should just learn that the timestamps
> have wrapped around without having to be told about it. He changed the
> code to work that way, so that full voice frames are only sent when the
> call is first set up, or when the timestamp has wrapped around more than
> once.

Nice enhancement - it went through my mind to try something like this;
I'm looking forward to testing it.  I think we need the extreme test
of two devices together on a LAN, talking via an Asterisk server 500
msecs away or something.

Steve Kann can comment further too - but my opinion is also that the
full voice frames don't serve much purpose any more.

We've gradually accumulated more and more situations where we are
depending on the receivers ability to "unwrap" the miniframe
timestamps.  So if we are dependent on that anyway, why bother with
the full frames.  (Unwrapping is not foolproof of course, but the
jitter buffer and other things are pretty resilient to timestamp
"funnies" too, see jumps can happen in various other situations).

Things to be checked, though:

1) Can we still detect in a reasonable time when the receiver has
"disappeared" at the IP level

2) Do we still interwork well with old IAX.  Probably need to go to
the 1.0-series releases?

Regards,
Steve