Le mercredi 14 juillet 2010 13:14:39, James Bensley a ?crit
:> You may be rich in bandwidth then but maybe you are experaincing
> packet loss? Can you test for this some how? (I can't recommend
> anything as I have never used Amazon EC2)
More than packet loss, a very likely possible explanation could be a clock
drift.
If the clock of the server, for instance, is too slow compared to the clock of
the listener, then the listener's buffer will drop to zero.
The popping/glitching sound could be the result of this buffer being to low and
the decoder trying to catchup.
The reconnection that reset the scenario could be just the initial buffer being
filled again.
This scenario is not theoretic, we have experienced it between a source client
and a liquidsoap harbor server. The only hacky solution we found was to
slightly drop the audio rate of the stream received in order to counter the
clock drift.
Unfortunately, this is probably not an option here because the clients can be
any audio player.
However, one way to test this issue is to use liquidsoap. The input.http
operator can log the length of its internal buffer, which can then be plotted.
If the clock drift assumption is correct, you should see a backward slope in
the plot.
Good luck,
Romain