Re: [Sipping] Call Completion: In Defense of Explicit Operations

[Adam Roach <[e-mail]>]

Francois:

You are correct that the specific inclusion of a requirement for strict 
call queuing to handle the once-in-a-lifetime situation in which two 
people attempt to perform call completion against the same destination 
is at the root of the complexity. Without this requirement, call 
completion services could be implemented using already-defined mechanisms.

Not only do I agree with your analysis; it's exactly what I was trying 
to convey at the Dallas meeting. If I recall, the pushback took the form 
of, "TISPAN has defined these requirements, and none of them are 
negotiable."

It would be nice if we could get some kind of liaison statement from 
TISPAN confirming that the use of strict queuing is considered a 
sufficiently critical requirement to warrant the relatively ridiculous 
complexity it introduces. (Relative, that is, to the impact on user 
experience.)

/a

Francois Audet wrote:
> Wouldn't it be possible to do this CCBS/CCNR stuff with a simple
> dialog package, and ditch this "queue management" mechanism?
>
> This seems to me to be a classic case of Engineers-gone-wild.
>
> I see this in the poetzl draft:
>
>     Completion of Calls on no Reply (CCNR) provides the caller with the 
>     ability to complete a requested call to a callee without having to 
>     make a new call attempt when the callee becomes not busy after 
>     having initiated an activity.  The CCNR service description is 
>     defined in ETSI EN 301 134 [6]. 
>
>     The call completion services provide the capability to queue 
>     several CCBS/CCNR requests to the same callee on a FIFO basis. It 
>     therefore requires functionalities like notifying changes of queue 
>     states and managing queues.  
>
> In other words, we are defining this really complex stuff just because
> it
> used to work like this in some ETSI standard for PSTN.
>
> Is there a better justification than that? What am I missing?
>
> Couldn't we handle some basic level of queueing at the endpoints
> themselves,
> and use a simple dialog package, and avoid all this stuff that creates
> all
> the complexity?
>
> This CCBS/CCNR feature is one of the feature that has the most variation
>
> in implementations in PSTN that I know of. To me, picking a really
> complicated
> variant and replicating it in SIP (either by bastardizing the protocol,
> or
> by using an entirely different protocol altogether) is not a good idea.
> My take is
> that there will be such a small uptake in implementation that it will
> effectively
> useful only in a few confined environments where the it can be mandated
> by law.
>
> I'd must rather have a simpler easier to implement version of the
> service.
>
>   
>> -----Original Message-----
>> From: Adam Roach [mailto:[e-mail]] 
>> Sent: Thursday, April 05, 2007 12:10
>> To: SIPPING WG
>> Subject: [Sipping] Call Completion: In Defense of Explicit Operations
>>
>> Despite what appeared to be early agreement in the working 
>> group (and on the MIG mailing list) that we wanted to 
>> discourage the use of SUBSCRIBE and NOTIFY as general purpose 
>> RPC mechanisms. This is the primary objection that I have 
>> raised to draft-poetzl-sipping-call-completion,
>> which uses SUBSCRIBE and NOTIFY to manipulate call queue state.
>>
>> In the spirit of sending text, I described an alternate set 
>> of mechanisms in draft-roach-sipping-callcomp-bfcp, which 
>> performs the same task without having implied side effects 
>> from the creation of a subscription.
>>
>> I'll further note that the use of SUBSCRIBE to both create a 
>> subscription and manipulate a call queue falls under the 
>> category of "one operation with two separate effects." We 
>> have a very poor institutional memory indeed if we cannot 
>> recall why operations of this nature are Really Bad Ideas 
>> (cf. REFER creating a subscription; REGISTER with call 
>> processing upload; and BYE with an "Also" header field).
>>
>> All of this notwithstanding, in Prague, there seemed to be a 
>> curious indifference to pursuing the path laden with such 
>> Really Bad Ideas. I suspect that this indifference is a 
>> result of the arguments raised so far against the mechanism I 
>> have proposed, which I have done little to refute so far 
>> (largely because I believed that the merits of a system with 
>> explicit operations were self-evident enough that the 
>> objections would seem trifling in comparison).
>>
>> As far as I can tell, the objections raised so far have 
>> fallen into three primary categories:
>>
>>    1. "The use of BFCP is troubling because, in a decomposed gateway,
>>       the BFCP goes to the Media Gateway, while the signaling goes to
>>       the Media Gateway Controller."
>>
>>       This objection, raised on one of the slides during the SIPPING
>>       meeting on March 23rd as the official reason TISPAN rejected the
>>       approach, is based on a simple misconception.
>>
>>       The BFCP goes wherever the SDP says the BFCP goes. In 
>> the case of
>>       a decomposed gateway, this could be the Media Gateway, the Media
>>       Gateway Controller, or any other random network-connected server
>>       that the Media Gateway Controller wants. I agree that 
>> it probably
>>       is most sensible to send it to the Media Gateway Controller, and
>>       this is trivial to achieve.
>>
>>    2. "In the case of PSTN interwork, there is no way to 
>> guarantee that
>>       the ISUP (or other) signaling from the PSTN side lands 
>> on the same
>>       gateway that the client has a BFCP connection to."
>>
>>       This is true, and it is a relatively difficult problem to solve
>>       (not impossible, though; I proposed several potential 
>> solutions in
>>       San Diego). It is meaningless to raise it in objection to an
>>       alternate proposal, since it exists in almost precisely the same
>>       form in draft-poetzl-sipping-call-completion; and any solution
>>       that can be applied to one solution can be applied to the other.
>>
>>    3. "We don't want to add another protocol."
>>
>>       I suspect this argument is the one that is receiving the most
>>       traction, probably because most of the people involved in the
>>       discussion are not familiar with BFCP. The protocol 
>> itself is very
>>       straightforward and extremely easy to implement. In fact,
>>       exclusively for the purpose of answering this email (and
>>       demonstrating this very point), I threw together a BFCP
>>       implementation sufficient for implementation of the Call
>>       Completion service described in 
>> draft-roach-sipping-callcomp-bfcp.
>>       It took me two hours, and compiles down to less than 4 kb of
>>       object code on an intel processor. It's attached to this message
>>       to help the participants in the conversation understand just how
>>       very little is being added to applications by this use of BFCP.
>>
>> So, is that it? The first two objections are simply red 
>> herrings, and the third is based on a misconception regarding 
>> the level of effort required to implement BFCP. Are we really 
>> going to down the path of doing things incorrectly over what 
>> amount to misunderstandings?
>>
>> /a
>>
>>
>>
>> P.S. In case anyone wants a more full-featured BFCP stack, 
>> I'll point to 
>> the confiance project on SourceForge, available under the Academic 
>> Freedom License, which is compatible with commercial development. See 
>> http://confiance.sourceforge.net/
>>
>>     


_______________________________________________
Sipping mailing list  https://www1.ietf.org/mailman/listinfo/sipping
This list is for NEW development of the application of SIP
Use [e-mail] for questions on current sip
Use [e-mail] for new developments of core SIP