Re: [MMUSIC] DTLS-over-SCTP, anyone?

[Christer Holmberg <[e-mail]>]

Hi Roman,

What about the following new section 3.2:

    “3.2.  Change of Local Transport Parameters

      If an endpoint modifies its local transport parameters (IP address
      and/or port), and if the modification requires a new DTLS
      association, the endpoint MUST either change its DTLS role, its
      fingerprint value and/or use the SDP 'dtls-connection' attribute with
      a 'new' value Section 4.

      If the underlying transport explicitly prohibits a DTLS
      association to span multiple transports, the SDP 'dtls-connection'
      attribute MUST be set to 'new' if the transport is changed.  An
      example of such case is when DTLS is carried over SCTP, as described
      in [RFC6083].”

Regards,

Christer

From: mmusic [mailto:[e-mail]] On Behalf Of Christer Holmberg
Sent: 11. helmikuuta 2016 9:09
To: Roman Shpount
Cc: [e-mail]
Subject: Re: [MMUSIC] DTLS-over-SCTP, anyone?

Hi Roman,

Your suggestion looks good. I’ll take it as a WGLC comment, and add it in the post-WGLC version of the draft :)

Regards,

Christer

From: Roman Shpount [mailto:[e-mail]]
Sent: 10. helmikuuta 2016 22:42
To: Christer Holmberg
Cc: [e-mail]<mailto:[e-mail]>
Subject: Re: [MMUSIC] DTLS-over-SCTP, anyone?


On Wed, Feb 10, 2016 at 2:58 PM, Christer Holmberg <[e-mail]<mailto:[e-mail]>> wrote:
>>> I think we can add the following to section 7.1 of dtls-sdp:
>>>
>>> If DTLS is transmitted over a reliable transport and if DTLS procedures for retransmissions are not used, for instance as described in
>>> RFC 6083 for DTLS over SCTP, then DTLS association MUST NOT span across multiple transports. Using 'dtls-connection' attribute with
>>> an 'existing' value in combination with change of such a reliable transport should be treated as an error and DTLS association MUST be
>>> terminated.
>>
>> Your text as such looks ok. But, do we really want to add it as a generic restriction in draft-dtls-sdp? Shouldn't it be
>> specific for DTLS-over-SCTP instead? What if someone defines DTLS-over-<new-fancy-reliable-transport> and they DO allow span?
>
> The reason multiple DTLS associations cannot span across several SCTP association is due to SCTP association handling DTLS packet
> retransmission and DTLS procedures for retransmissions not being used. We can make a generic statement or limit this to RFC 6083
> only, but I think, stating the reason why DTLS association cannot span multiple transports is important.

I agree.

But, again DTLS-over-SCTP is described in draft-sctp-sdp, so I think such text belongs there.


How about saying the following in draft-dtls-sdp:

If DTLS is transmitted over a transport that prohibits spanning of DTLS association across multiple transports, such as DTLS over SCTP as described in RFC 6083, then 'dtls-connection' attribute MUST be set to 'new' every time transport is changed.

And you can put the previous language in draft-sctp-sdp

This way each draft only specified things relevant to itself.

Regards,
_____________
Roman Shpount