[Asterisk-Dev] RTCP-'support'/jitter buffering, etc.

[Steve Kann]

Filip Olsson wrote:

>The solution to put these thin CQDR's in the userfield isn't at all the
>final solution, it was just a dirty hack to get them out.
>Although the userfield isn't the solution to this, I don't now if a
>separate record is the best one. Why not just put the stuff in the CDR
>but in separate fields?
>
>Concerning H.323/MGCP/SCCP it shouldn't be too much of a problem to get
>them to use this, although I don't really know because I'm not that
>familiar with them. But as long as they use the RTP from rtp.c the
>modifications should be minor.
>
>//filip
>
>On Mon, 2004-11-15 at 23:14, John Todd wrote:
>  
>
>>Wow, this is really great!   I'm warming up the RRDTool engine 
>>already to start plotting this data after I get the patch installed.
>>
>>Sorry to follow up so quickly with the RTCP XR stuff - I hadn't even 
>>read the list to see your message when I sent that.  :-)  The bad 
>>news is that RTCP-XR isn't in most equipment... yet.  The data you're 
>>collecting is excellent, though.
>>
>>My preference would be to avoid populating this data into the CDR 
>>field (even in the "user" field) since this repository is getting 
>>pretty crowded.  In fact, this would be ultimately also perfect for 
>>exportation into a database structure - the hooks for that exist 
>>already, right?  You've done 90% of the work for getting a "CQDR" 
>>data store running - any chance of getting out of the CDR log and 
>>then weaving h.323 and IAX2 into the mix?  Also, I'm not sure how 
>>you'd handle IAX2 - do the same RTP hooks apply?
>>
>>Can the developers of H.323/MGCP/SCCP speak to how this might be 
>>worked into those channels if/when this is more stable?
>>
>>JT
>>
>>
>>At 8:01 PM +0100 on 11/15/04, Filip Olsson wrote:
>>    
>>
>>>Hello boys and girls,
>>>
>>>I've posted a patch to
>>>http://bugs.digium.com/bug_view_page.php?bug_id=0002863 that add's some
>>>RTCP-support to rtp.c. It adds the ability to send/receive (and almost
>>>understand) RTCP sender/receiver reports.
>>>
>>>It's kind of hard to test if the receiver actually understands what we
>>>send them. The way I made sure it was correctly understood by the remote
>>>end was to make sure that such things as LSR and DLSR in RTCP RRs was
>>>filled in correctly to reflect our SR. The RTP-stack in 42Networks'
>>>DRG-series understand them and responds correctly, according to
>>>ethereal.
>>>
>>>The whole point of this patch was to dig out some statistics on the
>>>quality(delay/jitter/loss) of the RTP-streams and then dump it to
>>>somewhere useful. I've written a small patch to chan_sip so that some
>>>averages/maxs/mins are saved in the userfield of the CDR. John Todd has
>>>put forth a much better solution that is more generic(not just RTP)
>>>http://lists.digium.com/pipermail/asterisk-dev/2004-May/004180.html.
>>>But atleast it's a start to have the stuff in the CDR.
>>>It would be even more useful to(on a per call basis) be able to log a
>>>RTP-stream to a separate file that describes the whole stream/call in
>>>more detail so one can monitor the stats as the call goes on(for people
>>>that like fancy graphing).
>>>
>>>RTCP and RTP is described in RFC3550, it's the best documentation I've
>>>found.
>>>
>>>Anyways, please test the patch and give me some feedback. I'm in
>>>#asterisk/-dev/-bugs under my very secret nick folsson.
>>>
>>>Don't even think about patching your production systems with this one,
>>>it'll most certainly blow up your machine and get you fired.
>>>      
>>>

I'm actually starting to work on a replacement jitterbuffer for asterisk 
(see "new jitterbuffer" in the wiki), and it would, of course, have to 
be able to generate all of these statistics.  (The idea is also to make 
this generic enough that this jitterbuffer could be used to dejitter RTP 
as well as IAX).

My idea for reporting was to implement something like RTCP for IAX2;  At 
Astricon, Mark seemed open to the idea of tagging these statistics on 
to, e.g. the LAGRP frames, as additional IEs.

The same basic information that RTCP sends would be sent via IAX2 as well. 

-SteveK