Re: [MMUSIC] Mirja Kühlewind's Discuss on draft-ietf-mmusic-sctp-sdp-23: (with DISCUSS)

["Mirja Kuehlewind (IETF)" <[e-mail]>]

Hi again,

actually I have one more mostly editorial comment. I would probably recommend to not just say that the framing of rfc 4571 is used but instead specify the framing in this draft (where you oft course still can say that the framing is similar to 4571). The reason is that other than the framing format itself the rest of rfc 4571 is not relevant because you use SCTP on top.

And again, it would probably also be good to talk a little more about implication when you use SCTP on top of TCP, mostly regarding the two layers of congestion control that you get.

Mirja




> Am 19.02.2017 um 11:04 schrieb Mirja Kuehlewind (IETF) <[e-mail]>:
> 
> Hi Roman,
> 
> thanks again. And again I think more text is needed in the draft.
> 
> Mirja
> 
> 
>> Am 16.02.2017 um 18:39 schrieb Roman Shpount <[e-mail]>:
>> 
>> Hi All,
>> 
>> I think a little bit of background will help here.
>> 
>> UDP/DTLS/SCTP and TCP/DTLS/SCTP are designed to work with ICE (RFC 5245).
>> 
>> In ICE environments, during the nomination process, end points go through multiple candidate pairs, until the most preferred pair is found. During this selection process, data can be sent as soon as the first working pair is found, but the process still continues and candidate pairs can change while data is sent. Furthermore, if end points roam, for instance when mobile end point switches from mobile internet to wifi, end points will initiate an ICE restart, which will trigger a new nomination process between the new set of candidates and likely result in new nominated candidiate pair. When these candidates change, the same DTLS association continues to run, regardless whether it is running over udp or tcp candidate pair. Because of this, ICE tcp requires using RFC 4571 framing when sending data (https://tools.ietf.org/html/rfc6544#section-10.1). Otherwise, if TLS were used, a new TLS or DTLS session would be required every time candidate pair switches between tcp and udp candidates. In order to simplify transition between different underlying transports, DTLS is used for both udp and tcp candidates and TCP/DTLS/SCTP transport tag is defined to differentiate it from other protocols.
>> 
>> As far as TCP/DTLS/SCTP transport tag is concerned, please note that ICE end points are supposed to send a re-INVITE after nomination process is completed with the selected candidate address in the m= line. So, if tcp candidate is selected, re-INVITE must be sent with TCP/DTLS/SCTP transport tag in the m= line. Also, any offers/answers after the ICE nomination is complete, are supposed to send the currently selected candidate in the m= line, which will also be TCP/DTLS/SCTP in case tcp candidate is selected.
>> 
>> I hope this addresses your concern and explains why TCP/DTLS/SCTP is defined instead of TLS/SCTP.
>> 
>> Regards,
>> _____________
>> Roman Shpount
>> 
>> On Thu, Feb 16, 2017 at 10:24 AM, Christer Holmberg <[e-mail]> wrote:
>> Hi,
>> 
>> 
>> 
>> …
>> 
>> 
>> 
>>> The only question is what should appear in the m= proto line.
>> 
>> 
>> 
>> Even if nobody puts it in the m= line, I think it’s useful to keep it in the document, as it gives an overview how it’s realized etc.
>> 
>> 
>> 
>> It doesn’t cause any harm to keep it, and it’s “future proof” IF someone wants to use it without ICE at some point.
>> 
>> 
>> 
>> Regards,
>> 
>> 
>> 
>> Christer
>> 
>> 
>> 
>> 
>>> Am 16.02.2017 um 16:02 schrieb Eric Rescorla <[e-mail]>:
>>> 
>>> As Christer says. This design is optimized for making the media stack simpler, which
>>> using TLS here would not do.
>>> 
>>> -Ekr
>>> 
>>> 
>>> 
>>> 
>>> On Thu, Feb 16, 2017 at 7:00 AM, Christer Holmberg <[e-mail]> wrote:
>>> Hi,
>>> 
>>> ----------------------------------------------------------------------
>>> DISCUSS:
>>> ----------------------------------------------------------------------
>>> 
>>>>>> Why is this using TCP/DTLS/SCTP instead of TCP/TLS/SCTP?
>>>>>> 
>>>>> Because the way it is realized is by transporting SCTP on top of DTLS (as defined in draft-ietf-tsvwg-sctp-dtls-encaps) and
>>>>> transporting DTLS on top of TCP (defined in RFC 4571).
>>>> 
>>>> I got this but DTLS is a mapping to use TLS with UDP because UDP is an unreliable datagram transport. If you use TCP, you
>>>> should use TLS. And rfc4571 is not a mapping of DTLS to TCP.
>>> 
>>> The framing mechanism of RFC 4571 is used, with DTLS packets sent instead of RTP packets.
>>> 
>>> Regards,
>>> 
>>> Christer
>>> 
>>> 
>> 
>> 
>> 
>> 
>