Re: [Linphone-users] libosip

[Troy Cauble]

Ged Haywood wrote:

> Hi Simon,
>
> On Fri, 7 Nov 2003, Ged Haywood wrote:
>
>
> > I've now downloaded and installed > 0.12.1pre3 but I won't have time
> > to do any testing until next week.
> >
> > I'll let you know what I find.
>
>
> Bad news I'm afraid: 0.12.1pre3 is very much worse for us than 0.12.0
> in practically every way.  The only improvement is that it doesn't
> crash now, but there's more on that below.
>
> I can't get any sense out of sipomatic.  The recording of your voice
> sounds like a sick motor-boat.  It's the same whether the robot is on
> a local machine or a remote one.
>
> Although the calls are made and seem to connect properly, linphone
> grumbles about timestamps in the future and neither user can talk to
> the other.  There's no sound at all apart from the ringing tones.  The
> system clocks on the two machines are synchronised to UTC via ntpd so
> it's difficult to know how the timestamps could be wrong.

Just FYI,

RTP timestamps don't have anything to do with wallclock or machine time.
It's just a number incremented by the number of samples represented by
the packet.  With an 8000 samples per second rate and 10 msec packets,
the timestamp increments by 80 with each packet.  (This is not the same
as a byte count, unless you're using G.711.)

An RTP receiver typically uses a jitter buffer to reorder packets and
handle missing packets based on the RTP timestamps and sequence numbers.

I don't know anything about linphone's implementation in this area, but
you could use ethereal to see if the timestamp and sequence numbers are
incrementing correctly.  That would tell you whether the sender is sending
bad data or the receiver is mishandling it.

Both the "sick motor boat" description and the "timestamps in the future"
complaint are consistent with bad timestamps being sent or a broken jitter
buffer impl.

-troy