[asterisk-dev] [Code Review] Rework of T.38 negotiation and UDPTL API to address interoperability problems

[Steve Underwood]

Kevin P. Fleming wrote:
> Steve Underwood wrote:
>
>   
>> How do you interpret the meaning of T38FaxMaxDatagram? Its a stupid 
>> name, as nothing in T.38 is actually referred to as a datagram. They use 
>> the term packet for both the IFP and UDPTL levels.
>>
>> T.38 D.2.1.3.1 says its "The maximum size of the payload within an RTP 
>> packet that can be accepted by the remote device."
>>
>> T.38 D.2.3.5 says its "This parameter signals the largest acceptable 
>> datagram for the offering endpoint and the answering endpoint (i.e., the 
>> maximum size of the RTP payload). The answering endpoint may accept a 
>> larger or smaller datagram than the offering endpoint. Each endpoint 
>> should be considerate of the maximum datagram size of the opposite 
>> endpoint."
>>
>> What exactly is the RTP payload? Before adding redundancy, or after? 
>> What about UDPTL? They never mention T38FaxMaxDatagram as having any 
>> relevence to UDPTL, although most systems only support UDPTL and most 
>> send a value for T38FaxMaxDatagram. A transmitted RTP packet can 
>> obviously be larger than T38FaxMaxDatagram, as you need to add the 
>> framing words to the payload. What happens in the case of UDPTL, where 
>> the framing and redundancy coalesce?
>>
>> In practice it seems some systems treat T38FaxMaxDatagram as the maximum 
>> IFP length, and some treat it as the maximum UDPTL length. I infer this, 
>> because some system use a number too small for it to be anything but the 
>> maximum IFP length.
>>     
>
> The patch currently treats T38FaxMaxDatagram as the maximum UDPTL
> payload size. The application is then given a maximum IFP length
> computed by using the worst-case error correction overhead (and a little
> bit of UDPTL overhead) for the mode the session is currently using.
>
>   
>> I've always regarded the number sent to my software as kinda iffy, and 
>> just tried to keep my IFPs pretty short.
>>     
>
> I suspect that most stacks do that as well, since it's not well specified.
>
>   
>> This area really needs exploring more in the FoIP working group.
>>     
>
> Agreed. Do you want to bring it up, or shall I?
>
>   
I'll sort that one out.

Steve