[asterisk-dev] sip/rtp jitterbuffer in 1.4?

[Patrick]

On Mon, 2006-05-29 at 18:18 +0200, Roy Sigurd Karlsbakk wrote:
> >> the sip/rtp jitterbuffer from http://bugs.digium.com/view.php?id=3854
> >> has been in the works for more than a year, and has been in  
> >> testing from
> >> the last patch since 1.2.1 or so. My testing shows it makes G.711A  
> >> work
> >> well with crystal-clear audio even on an overloaded 704/128kbps link.
> >> since asterisk is quite unusable for large-scale itsp rollouts  
> >> without a
> >> jitterbuffer for RTP-based protocols, SIP in particular, it would be
> >> really nice to get slav's code into 1.4

I'd say it's a must have but to be nice I'll go with "nice" :)

> > Yes, it would be very nice.  But, have you even tested the code  
> > that is
> > the candidate for 1.4?  While testing the 1.2 branch of this code  
> > may be
> > useful to you, it does nothing for the project to help get it into  
> > 1.4

As has been said in other postings, there are many branches (or whatever
they are called). One simply can't test everything.

> > When Joshua Colp and I started testing this last week, we ran into
> > serious problems immediately.  When we tried the 1.2 version,  
> > everything
> > worked great.  If the trunk version gets into shape in the next few
> > days, it is going in for 1.4.

Well, why not put it in *trunk* instead of some other branch so it will
get a lot more exposure? For those not very familiar with subversion,
it's not even easy to find out how to get the branch that has the JB. If
you want testers you have to make it easy/easier for people to test the
code. SteveU's comments indicated that the opposite is/was the case.
That's hardly encouraging. Imho having to put in more effort will result
in less testing. When people have to fix stuff and thus need to be C
proficient, you will instantaneously loose all those who want to help by
testing and submitting reports but who know no/little C.

> > It's like Kevin said in another thread about this, "An unreliable,  
> > buggy
> > or unmaintainable jitter buffer is worse than none at all."

So stick it in trunk and let people try it out. If you want to you can
even use some kind of ifdef and disable this JB by default. 

> Can you at least please put it in there so people can start testing  
> it? Trunk code usually _is_ buggy, and I somehow doubt this code is  
> unmaintainable. According to zoa/slav the code is very thoroughly  
> tested on 1.2, so it should be no more than porting it to 1.4. They  
> may perhaps even have time to help.

I certainly hope that this is/will be the case.

> I know there was much argument about allowing the jitterbuffer into  
> 1.2 by release or not, and that it was decided against it. PLEASE do  
> not let this happen with 1.4. The PLC and jitterbuffer is is a  
> feature much missed in asterisk, and they're both in these works.

I recall those discussions and I would have appreciated the JB make 1.2.
The 1.2 release is already a long time ago and again I hope the JB makes
1.4. Fwiw I don't mind the 1.4 release slipping a bit if that means that
it comes with the JB. 

Regards,
Patrick