Andreas,

On Fri, 2008-01-11 at 15:41 +0100, Andreas Granig wrote: 
Let me clear this up. In some deployments billing relies on successful 
re-INVITEs (with an interval of some minutes) rather than on RTP 
timeouts (which may be detected within some seconds).
Right. That's the solution I like the most.

This works because 
they are rather closed systems with UACs which support the Session Timer 
standard.
Well, you only need 1 SIP endpoint in the chain supporting SST for this to work, but obviously if this one breaks, well, you loose it all :-)

So if a BYE is lost for some reason, the call is torn down on 
the next re-INVITE latest.
I know :-)

> I'm really curious, could you give me a real-world example of an
> RTP-detection based soution providing sub-minute dead UA detection ? 

No, sorry, since I also don't rely on RTP regarding the billing :)
Well, seems in the end we have the same approch :)

What was strange to me (and still is), is the fact that some people are using RTP detection with a few seconds threshold. This is insane in my environment, and really I don't see an environment where it would not be a huge penalty, for the reasons I've given before in this thread :
- you cannot use VAD
- you're stuck with RTP proxying
- users cannot cut the mike on their phone (with business subscribers, this is a clear no-go)
- it's not generic at all, aka, it does not work for SIP  sessions that are not phone calls
  (IM sessions, using something else than RTP, or just oneway RTP like message diffusion etc...)

So I'm saying it again: using RTP detection is an UGLY HACK (when used alone).
On the other hand, RTP detection + ping when you detect the stream went down, so you can check
if the stream went down but the user is still in control, is a good idea. So RTP detection is interesting
when used AS A PART OF A MORE COMPLEX SCHEME, but NOT standing alone.

What do you think ?

Regards,

Jérôme Martin | LongPhone
Responsable Architecture Réseau
122, rue la Boetie | 75008 Paris
Tel :  +33 (0)1 56 26 28 44
Fax : +33 (0)1 56 26 28 45
Mail : jmartin@longphone.fr
Web : www.longphone.com