[Serusers] [RFC] ideas about new dialog module

Maxim Sobolev sobomax at portaone.com
Fri Jan 17 12:44:42 CET 2003


Jan Janak wrote:
> 
> Hello,
> 
> comments inline.
> 
> On 16-01 23:37, Maxim Sobolev wrote:
> > Hi,
> >
> > I am thinking about writing a new module for SER, which will track SIP
> > dialogs and will serve as an abstraction layer for other modules, much
> > like the tm module now. We need such module for 2 reasons:
> 
>   We are working on dialog support. It is working already but still
>   not mature enough to release.

Is it in cvs? If not, how can I obtain it?

> > 1. Call accounting. Our billing engine is based on the assumption that
> > a node provides accounting information for completed calls, not for
> > individual transactions. It is easier for us to extend proxy with
> > similar features than to modify billing engine to do transaction
> > matching.
> 
>   You can match the INVITE and BYE transactions and you have the whole
>   call. IMHO it is easier to modify billing software to do this than
>   implement it in the proxy.
> 
>   From my point of view, dialog support in a proxy should be avoided
>   wherever possible. The dialog support includes some additional bottleneck
>   to the proxy. The proxy must maintain dialog state for the duration of
>   the call and that can be very memory intensive. Our tests show that even
>   transaction state can exhaust memory of an extremely loaded proxy in less
>   than a minute.

Well, since some entity (B2B UA) will have to maintain call state
anyway, this will only move botleneck to that entity, not eliminate
it. Therefore, I don't see why introducing that another entity and
additional level of indirection associated with it, when it is
possible to do it right in the proxy. After all, I am planning to do
it as a module, so that if somebody doesn't need this functionality he
will not load the module and have the full raw power of the ser at his
hands.

> > 2. Debit card application. Currently, there is no way to use SER for
> > debit card applications, where it is necessary to set the maximum
> > duration of the call and terminate it forcefully if that duration is
> > exceeded.
> 
>   Yes, there is no way to do it, that's a feature, not bug. Dialog support
>   is not enough to make this possible. To be able to terminate a call, you must
>   implement a back to back user agent. A back to back user agent is an entity
>   that maintains two dialog states and acts as a UAS on one side and UAC on
>   the other. That means it terminates calls on one side and initiates call
>   on the other side. So even if you had dialog support in ser, you would be
>   unable to terminate calls.

I disagree - who prohibits you to send two BYEs to both parties once
the timer is expired impersonating itself as UAS for one party's UAC
and UAC for another party's UAS? Since you was in the middle of the
call flow, you obviously have every piece of information necessary for
doing it.

>   Ser has been designed as a proxy, not a user agent.
> 
> > The raw idea is as follows:
> >
> > - the module will register callbacks with tm.register_tmcb(), probably
> > TMCB_REQUEST_IN and TMCB_REPLY_IN ones and will match INVITEs to BYEs
> > keeping information about the state of ongoing sessions in the shared
> > memory.
> >
> > - the module will provide interested modules with ability to register
> > several callbacks, i.e. on dialog creation/teardown and yet another
> > callback on dialog timeouts (more about that below).
> >
> > - the module will provide utility functions for forceful termination
> > of any ongoing dialog.
> 
>   How do you want to terminate a dialog if you are an endpoint
>   of the dialog ?

See above.

-Maxim



More information about the sr-users mailing list