Re: [bfcpbis] BFCPbis: UDP- and TCP candidates and proto value

[Christer Holmberg <[e-mail]>]

Hi,

>>>>While the text is fine and probably closest to current behaviour, it
>>>>means we can¹t get the benefits from ICE if one end supported only one
>>>>of UDP or TCP.
>>>>
>>>> Today if you offer only TCP or UDP as the proto value on the m-line,
>>>>and the far end doesn¹t, then the far end will zero-port the m-line and
>>>>you¹ll 
>>>> either re-INVITE if you support the other, or just run without BFCP.
>>>>That same behaviour would continue with the proposed text.
>>>>
>>>> However, the real benefit of ICE in this circumstance would allow the
>>>>offerer to offer TCP
>>>> and UDP candidates under a TCP m-line, but the answerer, if they
>>>>supported only UDP, could then only offer back only UDP
>>>>
>>>>
>>>> candidates (whilst matching the offerer¹s proto). But given the text
>>>>requires a default candidate to be present matching the proto on the
>>>>m-line, then if the answerer did not support TCP, they would have to
>>>>zero-port the line.
>>>>
>>>> However, I would assume that breaking the semantics of the offered
>>>>proto field is probably not a good thing, and defining a
>>>>transport-agnostic proto isn¹t practical, so borrowing the proposed
>>>>text (with the photo -> proto typo fixed) is probably
>>>>
>>>> sufficient at this stage. Unless anyone can think of a neater way of
>>>>solving this?
>>>
>>>
>>> I guess one solution would be to allow the answerer to use a m- line
>>>proto value that does NOT match the default candidate (or, doesn¹t match
>>>ANY candidate).
>>
>>
>> That would certainly work in this scenario - different from the SCTP
>>text, but would permit this behaviour, whilst still providing clear
>>guidance.
>
>We would update the SCTP text too.
>
>> However, I fear that would go against the ICE spec; specifically, 5245
>>says:
>>
>>   The transport addresses that will be the default destination for
>>   media when communicating with non-ICE peers MUST also be present as
>>   candidates in one or more a=candidate lines.
>>
>> So we¹d no longer be adhering to that in the answer.
>
>The text is certainly valid for the offer, but when the answer is sent it
>is known whether the peers support ICE or not.
>
>In any case, I don¹t think there should be different rules for BFCF, SCTP
>etc. This should be defined in draft-ietf-mmusic-ice-sip-sdp as a generic
>rule.


I was thinking a little more about this: maybe indicating a transport in
the m- line that you don’t support isn’t a very good idea - even if it
won’t be used with ICE.

Maybe it would be better to say that the m- line shall contain a transport
that the peer is “most likely” to support. In case of BFCP, I guess
neither TCP or UDP is mandatory to support, but in other cases there is
often a mandatory transport.

Regards,

Christer
>