[MMUSIC] Proto identifier: To UDP or not to UDP (Re: Review of draft-ietf-mmusic-sctp-sdp-03)

[Harald Alvestrand <[e-mail]>]

One high level comment:

This series of comments points out that the purpose of the thing called 
"proto identifier" in SDP is not really well documented.

As far as I can tell, the requirement is that it uniquely identifies the 
parser one needs to use for the stuff under this m= line, and that is 
the sum total of the requirements on it - as we painfully learned with 
RTP/AVP, RTP/AVPF, SRTP/AVP and SRTP/AVPF, making detailed information 
about the parameters available via the "proto identifier" is not really 
a Good Idea.

I suggest we go for DTLS/SCTP, as specified in the document currently, 
and document that the lower layer used has to be identified from the 
elements inside the m= line block; if there is ICE, ICE can negotiate 
it; if there is only a port number, it's UDP; if there are more 
extensions to specify that the lower layer is LTE over pterodactyls, the 
identifier doesn't change.

Same handling as for RTP/AVP(F), and no double specifications.

On 03/20/2013 04:11 PM, Magnus Westerlund wrote:
> Hi,
>
> I have reviewed the SDP description of SCTP and have some comments and
> suggestions.
>
> 1. My first comments is a very general one and concerns the relationship
> between the proposed SCTP related proto identifiers and ICE for the
> DTLS/SCTP encapsulation.
>
> When ICE was defined it was natural to include ICE by reftorfitting it
> to existing proto identifiers. However, given that this specification
> intorduce a number of new identifiers I do woner if we should be more
> explicit about the ICE usage in the proto field. I think have both pro
> and cons for this.
>
> So my considerations is basically to include specification of
> ICE/DTLS/SCTP, where the lower layer transport is defined by what ICE
> agrees on. This makes a lot of sense given that ICE can in fact arrive
> on the usage of TCP/framing/DTLS/SCTP, TCP/TURN/DTLS/SCTP in addition to
> more expected choices as UDP/DTLS/SCTP. This makes it clears that you do
> require ICE to negotiate this transport flow. At the same time it is a
> significant downside if ones negotiation needs are to chose between
> using ICE or a specific lower layer transport.
>
> These comments doesn't apply when one runs SCTP on top of IP.
>
> 2. Section 1:
> RFC4145 [RFC4145] specifies a general mechanism for
>     describing and establishing TCP (Transmission Control Protocol)
>     streams.
>
> I do dislike that usage of the RFC number as the explanation of what the
> referenced document is. The previous sentence is how I prefer it to be
> done:
>     SDP (Session Description Protocol) [RFC4566] provides a general-
>     purpose format for describing multimedia sessions in announcements or
>     invitations.
>
> Use the document title or an abbreviated explanation for what the
> referenced document specifies and one gets much better flow in the
> reading as one don't have stop look up the reference to find out what is
> referred to.
>
> 3. Section 1:
> I think having first a paragraph for SCTP and then having "Additionaly
> .." with the other two protos are bit strange. Why not simpley say. This
> document defines three proto identifiers: list of them with explnation.
>
> 4. Section 1:
>     DTLS/SCTP :  to allow the usage of SCTP on top of the Datagram
>        Transport Layer Security (DTLS) protocol, as defined in
>        [I-D.tuexen-tsvwg-sctp-dtls-encaps], using SDP.  SCTP over DTLS is
>        used by the RTCWeb protocol suite for transporting non- media data
>        between browsers.
>
> This definitions says nothing of the lower transport layer for DTLS/SCTP
> even if UDP is referred to in a later section. I think this needs to be
> clarified and discussed in the context of which lower layers do we
> expect to be supported? From my perspective I think two makes sense.
> ICE/DTLS/SCTP that allows ICE to negoiate what ever and then
> UDP/DTSL/SCTP that only uses UDP as lower layer. Lets other add other
> variants in the future if needed.
>
> 5. Section 3:
>     The 'DTLS/SCTP' protocol identifier indicates that the media
>     described will use SCTP on top of the Datagram Transport Layer
>     Security (DTLS) protocol as specified in
>     [I-D.tuexen-tsvwg-sctp-dtls-encaps].
>
> I would like to point out that the referenced document does not at all
> indicate what lower layer is being used under DTLS.
>
> 6. Section 4.1:
>
> This section discuss the usage of the "data channels" within the SCTP
> association. My personal position is that for the moment this appears to
> be unnecessary. The most important part is the SCTP association
> establishment. Then one can discuss the general application using the
> SCTP association as whole. Examples of such are WebRTC data channel.
>
> If anyone want stream level information in SDP then I propose that this
> is handled as a extension to this signaling, not an from the start
> included functionality as we don't appear to have clear requirement for
> that.
>
> 7. Section 4.2:
>
> What are the requirements behind being able to establish multiple SCTP
> association over the same DTLS connection? I am very unclear why this
> would be required, and if not really needed I would suggest keeping
> things simple.
>
> 8. Section 4.3: I guess this can for the moment be removed as it appears
> to be taken care in other places, such as the RTCWEB WG data channel
> protocol proposal.
>
> 9. If you are removing the multiple association and the "data channel"
> parameters then most of the issues in section 5 vanish. If not I have
> additional feedback on this section.
>
> 10. Section 8.
>
>     Current NATs do not typically support SCTP.  As an alternative to
>     design SCTP specific NATs, Encapsulating SCTP into UDP
>     [I-D.tuexen-sctp-udp-encaps] makes it possible to use SCTP in
>     networks with legacy NAT and firewalls not supporting SCTP.
>
> First of all, there is a WG version for this document:
> draft-ietf-tsvwg-sctp-udp-encaps-14. Which is in IESG evaluation. Thus I
> wonder why signalling for this is not included when all the other
> options are included, even less mature ones then this?
>
> To me it looks like this document need to have some high level decisions
> on what to include and not and then try to finish it up within the
> agreed constraints. From my perspective I think it should focus on a
> single SCTP association per m= line and the actual SCTP association.
>
> Cheers
>
> Magnus Westerlund
>
> ----------------------------------------------------------------------
> Multimedia Technologies, Ericsson Research EAB/TVM
> ----------------------------------------------------------------------
> Ericsson AB                | Phone  [phone]
> Färögatan 6                | Mobile [phone]
> SE-164 80 Stockholm, Sweden| mailto: [e-mail]
> ----------------------------------------------------------------------
>
> _______________________________________________
> mmusic mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/mmusic