| rtcweb | D. Worley |
| Internet-Draft | Ariadne |
| Intended status: Standards Track | March 11, 2013 |
| Expires: September 12, 2013 |
Kumquat: A Generic Bundle Mechanism for the Session Description Protocol (SDP)
draft-worley-sdp-bundle-05
This document defines a generic bundle mechanism for the Session Description Protocol (SDP) by which the media described by a number of media descriptions ("m= lines") are multiplexed and transmitted over a single transport association. The transport association is described by an additional media description, allowing SDP attributes to be applied to the aggregate, independently of attributes applied to the constituents. In offer/answer usage, the bundle mechanism is backward compatible with SDP processors that do not understand the mechanism. The mechanism is designed to be compatible with the limitations of the existing Internet infrastructure.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on September 12, 2013.
Copyright (c) 2013 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
The central idea of bundling is to multiplex the media that would be several RTP sessions into one RTP session, with particular emphasis on allowing one transport association to carry media that are presented to the higher, application layer, as multiple RTP sessions.
At the interface between the SDP-configured layer and the lower, transport layer, the media are organized into a single RTP session. The transport-related properties of the RTP session (e.g., transport 5-tuple, encryption, ICE) are described by the transport-related attributes of a single media description.
At the interface between the SDP-configured layer and the higher, application layer, the media are organized into several RTP sessions. The application-related properties of the RTP session (e.g., media type and label) are described by the application-related attributes of separate media descriptions.
(There are some attributes (e.g., bandwidth limitation) that can apply separately to both the bundled RTP session and the constituent RTP sessions.)
However, we do not include the payload type numbers as information available to the application; only the encoding name and its parameters are accessible to the application. This gives the bundle mechanism freedom to place constraints on the use of payload types.
The bundle is signaled in the session description by a "group" attribute with semantics "KUMQUAT". The first media description listed in the group is the "bundle" media description (MD), whose transport information describes the transport association via which the RTP packets will be sent. The remaining (zero or more) media descriptions listed in the group are the "constituent" MDs. RTP packets received from the applications for these MDs are encapsulated and sent on the transport association for the bundle MD. RTP packets received from the transport association for the bundle MD are deencapsulated and sent to the applications for the constituent MDs.
A new payload type (codec) named "kumquat" is defined to be used for this encapsulation. Section 5.4.1
In offer/answer usage, we must arrange that the bundle mechanism is backward compatible with entities that do not understand the bundle mechanism. This requirement drives many features of this solution. Section 6.1
In addition, many devices in current usage (especially SBCs) apply more restrictions on the usage of SDP than one would expect from abstract consideration of their roles in the network. Some features of this solution are constructed to avoid these restrictions. Section 6.2
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].
The important RFCs in this area use inconsistent terminology. Here, we use:
This section lists desiderata for the bundle mechanism in SDP. (I use the term "desiderata" -- "things that are desired" -- rather than "requirements", because we may discover that we can't optimally satisfy all of these criteria at the same time.) The first section lists desiderata that are arise from considering the ways applications may wish to bundling. The second section lists desiderata that arise from compatibility with existing Internet infrastructure.
These desiderata describe features that we would like the bundling mechanism to provide.
This requirement is taken from [I-D.jennings-mmusic-media-req].
DES F1 and DES F2 do not specify whether the transport-level media description may or may not also be one of the application-level media descriptions.
This desideratum is taken from slides-interim-2013-rtcweb-1-10.pdf.
Of course, no bundle may directly or indirectly contain itself. (I don't expect any current implementation to implement bundles within bundles, but we should design the mechanism to allow this, as some day we will likely need it.)
A bundle with no constituents serves no purpose for the transport of media, but we are likely to someday need to describe such a bundle. (Compare that an SDP m= line is syntactically constrained to specify at least one payload type. When SDP was used only to specify multicast sessions, this constraint was common sense. But once SDP offer/answer was invented, when a media description was rejected, the natural representation would be an m= line with a zero port and no payload types. But a payload type was syntactically required, so we now have to provide at least one token payload type in rejected m= lines.)
Presumably answer (3) resembles that which would be produced by an answerer that does not understand the bundle mechanism. It is a lower priority that the answerer can distinguish between accepting the bundle while rejecting all of its constituents, and rejecting the bundle as a whole. But those two conditions differ conceptually regarding whether any "framing" actions of the bundle are performed.
The RTCP information for each media stream is tagged with the SSRC about which it reports, and the SSRC is used to correlate the RTCP reports with the RTP sessions containing media with the same SSRC. So regarding RTCP, this desideratum appears to be straightforward to satisfy.
In the terminology of [RFC3550], the constituent media descriptions are now part of one RTP session.
This desideratum was suggested by Andrew Hutton.
Presumably this can be accomplished as it is now, with a single media description carrying multiple video flows that are distinguished only by their SSRCs. This desideratum is taken from slides-interim-2013-rtcweb-1-10.pdf.
This desideratum was suggested by Andrew Hutton.
These desiderata describe compatibility of the bundling mechanism with with non-supporting endpoints or with existing entities in the Internet infrastructure.
SDP features (e.g., the codec set and ICE) are generally designed so that an offerer always offers every facility it is willing to support in the current situation, regardless of whether it was agreed to by the answerer in a preceding exchange. Thus, if the current answerer is a different endpoint than the previous answerer, the new answerer will negotiate a compatible set of facilities without needing knowledge of its predecessor's SDP. The offerer will smoothly transition to the new facilities. This property is required to support 3PCC situations (e.g., [RFC3725] and [I-D.worley-service-example]). This desideratum was suggested by Richard Ejzak.
This desideratum was suggested by Hadriel Kaplan.
Many devices that have only one audio or video channel accept the first m= line with that media type and reject any further ones
This non-desideratum was suggested by Hadriel Kaplan.
This is needed because SBCs monitor the packet traffic on the transport associations and if no media is seen on one of the associations for a significant period of time, the SBC will tear down the call. This desideratum was suggested by Hadriel Kaplan.
Such duplication is not defined by [RFC4566]. Some SBCs do not support such duplication (ultimately, because it was not supported by [RFC2327]), and they reject SDP specifying duplicated transport association endpoints. This desideratum was suggested by Cullen Jennings.
The non-encrypted case is not expected to be very common. Encrypted media can't be transcoded by an intermediate entity.
This section is non-normative. (This section was suggested by Charles Eckel.)
This is an introduction to SDP bundling via a series of examples of offer/answer processing. Some mandatory SDP lines have been omitted from the examples for brevity. Long SDP lines have been folded by using trailing backslashes. Blank lines have been inserted for clarity.
Here is a typical, non-bundled SDP example with both audio and video media:
o=- 2890844526 2890844526 IN IP4 host.example.com
c=IN IP4 10.0.1.1
This SDP media description (MD) provides the transport information
about the audio and also identifies the role of the audio from the
application's point of view. In this case, the fact that it is the
first audio m= line suffices to tell the application how to treat it.
In more complex cases, label or content attributes might be used to
communicate the proper handling to the application.
m=audio 10000 RTP/AVP 0 8 97
a=rtcp-mux
a=rtpmap:0 PCMU/8000
a=rtpmap:8 PCMA/8000
a=rtpmap:97 iLBC/8000
a=candidate:0 1 UDP 2113601791 10.0.1.1 10000 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.32 51000 typ srflx \
raddr 10.0.1.1 rport 10000
This MD provides the transport information about the video and also
identifies the role of the video from the application's point of view.
m=video 10002 RTP/AVP 31 32
a=rtcp-mux
a=rtpmap:31 H261/90000
a=rtpmap:32 MPV/90000
a=candidate:0 1 UDP 2113601791 10.0.1.1 10002 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.32 51002 typ srflx \
raddr 10.0.1.1 rport 10002
We call the RTP that is described by each media description (MD) a transport flow (TF). The audio and video are carried in separate TFs, which each have a separate transport association (address/port).
With SDP bundling, we add an additional MD to describe a single "bundle" TF to carry both the audio and video information, and a group attribute to show the association of the bundle MD with the constituent MDs:
o=- 2890844526 2890844526 IN IP4 host.example.com
c=IN IP4 10.0.1.1
The following group attribute declares which MDs are included in the
multiplexed MD: mid:con1 and mid:con2 are the constituent MDs whose
TFs (from the application point of view) will be carried by the TF of
the first-designed MD, mid:bundle, which is the bundle MD.
a=group:KUMQUAT bundle con1 con2
This MD provides the application-level description of the audio TF.
As in the previous example, it is the first audio m= line. It
includes any attributes which apply to the audio media from the
application point of view, including the payload type definitions.
When interpreted by a supporting processor, the transport information
is ignored. When interpreted by a processor that does not support
bundling, the transport information sets up the transport association
for the audio TF.
m=audio 10002 RTP/AVP 0 8 97
a=mid:con1
a=rtcp-mux
a=rtpmap:0 PCMU/8000
a=rtpmap:8 PCMA/8000
a=rtpmap:97 iLBC/8000
This MD provides the application-level description of the video TF.
As in the previous example, it is the first video m= line. It
includes any attributes which apply to the video media from the
application point of view. As in the audio MD, the association
information is used only by a processor that does not support
bundling.
m=video 10004 RTP/AVP 31 32
a=mid:con2
a=rtcp-mux
a=rtpmap:31 H261/90000
a=rtpmap:32 MPV/90000
This MD provides the transport information for the bundle TF,
including any attributes which apply to the transport. We use RTCP
multiplexing [RFC5761], so only one set of ICE candidates (and only
one TURN relay) is needed for each MD. The MD is artificially given
the media type "audio" (which is ugly, but it avoids rejection by
SBCs) and it is placed after all of the constituent MDs so as to not
affect their positions as "first audio MD", etc. The MD lists a
single payload type for the "kumquat" payload format, which is used to
encapsulate the RTP of the constituent TFs.
m=audio 10000 RTP/AVP 127
a=mid:bundle
a=rtcp-mux
a=rtpmap:127 kumquat
a=candidate:0 1 UDP 2113601791 10.0.1.1 10000 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.32 51000 typ srflx \
raddr 10.0.1.1 rport 10000
If this SDP bundle is accepted, RTP provided by the application for the audio TF will be encapsulated into a kumquat payload and then be sent from port 10000. The encapsulation also contains the ordinal index (i.e., 0) of the audio TF and the payload type of the original audio RTP. RTP provided by the application for the video TF will be encapsulated into a kumquat payload and then be sent from port 10000. The encapsulation also contains the ordinal index (i.e., 1) of the video TF and the payload type of the original video RTP.
RTP that is received on port 10000 is interpreted according to the kumquat payload format: The constituent MD ordinal index is extracted. The encapsulated RTP and its payload type are then interpreted according to the constituent MD.
If the answerer supports SDP bundling, and desires to accept the offered bundle and its constituent MDs, the answerer signals that it accepts the SDP bundling by providing a matching group:KUMQUAT attribute in the answer. As always in offer/answer, the MDs in the answer correspond to the MDs in the offer by ordinal position.
The answerer provides the necessary transport information for the bundle MD. The answerer understands that MDs mid:con1 and mid:con2 are incorporated into MD mid:bundle, and ignores their transport information. It accepts each constituent MD by providing an answer MD for each of them that specifies a null address and port 9 (the discard port).
o=- 2890844526 2890844526 IN IP4 answer.example.com
c=IN IP4 10.0.2.1
a=group:KUMQUAT bundle con1 con2
m=audio 9 RTP/AVP 0 8 97
c=IN IP4 0.0.0.0
a=mid:con1
a=rtcp-mux
a=rtpmap:0 PCMU/8000
a=rtpmap:8 PCMA/8000
a=rtpmap:97 iLBC/8000
m=video 9 RTP/AVP 31 32
c=IN IP4 0.0.0.0
a=mid:con2
a=rtcp-mux
a=rtpmap:31 H261/90000
a=rtpmap:32 MPV/90000
m=audio 20000 RTP/AVP 127
a=mid:bundle
a=rtcp-mux
a=rtpmap:127 kumquat
a=candidate:0 1 UDP 2113601791 10.0.2.1 20000 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.35 51090 typ srflx \
raddr 10.0.2.1 rport 20000
Because the offer contained real addresses and ports for the constituent MDs and the answer accepted them, intermediate entities may expect the offerer to send at least RTCP on the transport association. To prevent such an intermediate entity from timing-out the multimedia session (because such RTCP will not be sent), the offerer needs to update its offer to withdraw the real address and ports for the constituent MDs, replacing them with a null address and port 9.
o=- 2890844526 2890844527 IN IP4 host.example.com
c=IN IP4 10.0.1.1
a=group:KUMQUAT bundle con1 con2
m=audio 9 RTP/AVP 0 8 97
c=IN IP4 0.0.0.0
a=mid:con1
a=rtcp-mux
a=rtpmap:0 PCMU/8000
a=rtpmap:8 PCMA/8000
a=rtpmap:97 iLBC/8000
m=video 9 RTP/AVP 31 32
c=IN IP4 0.0.0.0
a=mid:con2
a=rtcp-mux
a=rtpmap:31 H261/90000
a=rtpmap:32 MPV/90000
m=audio 10000 RTP/AVP 127
a=mid:bundle
a=rtcp-mux
a=rtpmap:127 kumquat
a=candidate:0 1 UDP 2113601791 10.0.1.1 10000 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.32 51000 typ srflx \
raddr 10.0.1.1 rport 10000
The answerer responds with the same answer as before.
SDP bundling allows for backward compatibility in case the answerer does not understand bundling. If the answerer does not understand bundling, it ignores the group attribute, and effectively sees the offer as:
o=- 2890844526 2890844526 IN IP4 host.example.com
c=IN IP4 10.0.1.1
m=audio 10002 RTP/AVP 0 8 97
a=rtcp-mux
a=rtpmap:0 PCMU/8000
a=rtpmap:8 PCMA/8000
a=rtpmap:97 iLBC/8000
m=video 10004 RTP/AVP 31 32
a=rtcp-mux
a=rtpmap:31 H261/90000
a=rtpmap:32 MPV/90000
m=audio 10000 RTP/AVP 127
a=rtcp-mux
a=rtpmap:127 kumquat
a=candidate:0 1 UDP 2113601791 10.0.1.1 10000 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.32 51000 typ srflx \
raddr 10.0.1.1 rport 10000
If the answerer wishes to accept the first audio and video streams, it assembles this answer:
o=- 2890844526 2890844526 IN IP4 answer.example.com
c=IN IP4 10.0.2.1
The absence of the group attribute informs the offerer that bundling
was rejected.
The audio MD is accepted. Transport information is provided, but it
does not include ICE candidates, because the offer did not provide ICE
candidates for the first and second MDs.
m=audio 20000 RTP/AVP 0 8 97
a=rtcp-mux
a=rtpmap:0 PCMU/8000
a=rtpmap:8 PCMA/8000
a=rtpmap:97 iLBC/8000
The video MD is accepted. Transport information (using a different
port) is provided.
m=audio 20002 RTP/AVP 31 32
a=rtcp-mux
a=rtpmap:31 H261/90000
a=rtpmap:32 MPV/90000
The bundle MD is rejected by the answerer because the only offered
codec was kumquat, and the answerer does not implement it.
m=audio 0 RTP/AVP 127
Because the group attribute is not present in the response, the offerer knows that the answerer does not support bundling (or does not want to consider the offered bundle). The offerer knows that the answerer wants to establish one audio TF and one video TF, and formally, that has been done. But if transport requires ICE candidates describing TURN relays, the offerer must send an updated offer containing those ICE candidates for the constituent MDs:
o=- 2890844526 2890844527 IN IP4 host.example.com
c=IN IP4 10.0.1.1
No group attribute is included, to ensure that this update only sets
transport attributes, and does not trigger bundle-supporting behavior
if the answering entity has changed in the meantime.
Provide ICE candidates for the audio MD. (We can reuse the ICE
candidates (and TURN relay) previously offered for the bundle MD.)
m=audio 10000 RTP/AVP 0 8 97
a=mid:con1
a=rtcp-mux
a=rtpmap:0 PCMU/8000
a=rtpmap:8 PCMA/8000
a=rtpmap:97 iLBC/8000
a=candidate:0 1 UDP 2113601791 10.0.1.1 10000 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.32 51000 typ srflx \
raddr 10.0.1.1 rport 10000
Provide ICE candidates (and a separate TURN relay) for the video MD.
m=video 10002 RTP/AVP 31 32
a=mid:con2
a=rtcp-mux
a=rtpmap:31 H261/90000
a=rtpmap:32 MPV/90000
a=candidate:0 1 UDP 2113601791 10.0.1.1 10002 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.32 51002 typ srflx \
raddr 10.0.1.1 rport 10002
The bundle MD must still be listed, but it is disabled.
m=audio 0 RTP/AVP 127
a=mid:bundle
The answerer then provides an answer that contains ICE candidates:
o=- 2890844526 2890844527 IN IP4 answer.example.com
c=IN IP4 10.0.2.1
m=audio 20000 RTP/AVP 0 8 97
a=rtcp-mux
a=rtpmap:0 PCMU/8000
a=rtpmap:8 PCMA/8000
a=rtpmap:97 iLBC/8000
a=candidate:0 1 UDP 2113601791 10.0.2.1 20000 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.35 51090 typ srflx \
raddr 10.0.2.1 rport 20000
m=audio 20002 RTP/AVP 31 32
a=rtcp-mux
a=rtpmap:31 H261/90000
a=rtpmap:32 MPV/90000
a=candidate:0 1 UDP 2113601791 10.0.2.1 20002 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.35 51092 typ srflx \
raddr 10.0.2.1 rport 20002
m=audio 0 RTP/AVP 127
The ICE negotiations proceed, the transport associations are established, and RTP flows.
The baseline procedure requires the offerer to update its offer when it discovers that the answerer does not support SDP bundling if TURN relays are needed to support the constituent MDs. The offerer can avoid this delay by providing transport information for the constituent MDs as well as for the bundle MD. The penalty is that the offerer must preallocate TURN relays for both the constituent MDs as well as the bundle MD.
o=- 2890844526 2890844526 IN IP4 host.example.com
c=IN IP4 10.0.1.1
a=group:KUMQUAT bundle con1 con2
m=audio 10000 RTP/AVP 0 8 97
a=mid:con1
a=rtcp-mux
a=rtpmap:0 PCMU/8000
a=rtpmap:8 PCMA/8000
a=rtpmap:97 iLBC/8000
a=candidate:0 1 UDP 2113601791 10.0.1.1 10000 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.32 51000 typ srflx \
raddr 10.0.1.1 rport 10000
m=video 10002 RTP/AVP 31 32
a=mid:con2
a=rtcp-mux
a=rtpmap:31 H261/90000
a=rtpmap:32 MPV/90000
a=candidate:0 1 UDP 2113601791 10.0.1.1 10002 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.32 51002 typ srflx \
raddr 10.0.1.1 rport 10002
m=audio 10004 RTP/AVP 127
a=mid:bundle
a=rtcp-mux
a=rtpmap:127 kumquat
a=candidate:0 1 UDP 2113601791 10.0.1.1 10004 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.32 51004 typ srflx \
raddr 10.0.1.1 rport 10004
If the answerer understands bundling and accepts the bundle, it accepts the constituent MDs (with a null address and port 9) and accepts the bundle MD. If the answerer does not understand bundling, it accepts the constituent MDs and rejects the bundle MD.
In this example, a presentation involves four media roles: the speaker's audio, the floor microphone, the video of the speaker, and the video of the speaker's slides. We use separate MDs for each media stream because each TF has a different role; the application will handle each of them in distinctly different ways.
o=- 2890844526 2890844526 IN IP4 host.example.com
c=IN IP4 10.0.1.1
a=group:KUMQUAT b c1 c2 c3 c4
m=audio 10002 RTP/AVP 0 8 97
a=mid:c1
a=label:speaker-audio
a=rtcp-mux
a=rtpmap:0 PCMU/8000
a=rtpmap:8 PCMA/8000
a=rtpmap:97 iLBC/8000
Note that different constituent MDs can use the same payload types
(for the same or different codecs), because the kumquat encapsulation
captures the constituent MD ordinal index separately from the payload
type.
m=audio 10004 RTP/AVP 0 8 97
a=mid:c2
a=label:floor-mic
a=rtcp-mux
a=rtpmap:0 PCMU/8000
a=rtpmap:8 PCMA/8000
a=rtpmap:97 G722
m=video 10006 RTP/AVP 103 104
a=mid:c3
a=label:speaker-video
a=rtcp-mux
a=rtpmap:103 H261/90000
a=rtpmap:104 MPV/90000
m=video 10008 RTP/AVP 103 104
a=mid:c4
a=label:slides
a=rtcp-mux
a=rtpmap:103 H261/90000
a=rtpmap:104 MPV/90000
m=multipart 10000 RTP/AVP 127
a=mid:b
a=rtcp-mux
a=rtpmap:127 kumquat
a=candidate:0 1 UDP 2113601791 10.0.1.1 10000 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.32 51000 typ srflx \
raddr 10.0.1.1 rport 10000
This example is the teacher's connection to a virtual classroom server. The media descriptions are tagged using the "content" attribute. [RFC4796] The media comprises:
1. one audio channel, for sending the teacher's voice and receiving the voice of a selected student
2. one video channel, for sending the teacher's presentation
3. one video channel, for sending the teacher's face
4. one video channel, for receiving a dynamically varying set of students' faces
The fourth TF (for students' faces) contains a large and dynamically varying set of video captures. These can be handled by a single TF because they all have essentially similar roles -- the application will process them as a set. As Adam Roach would say, "no control surfaces are necessary to talk about and/or manipulate the individual streams". In particular, this allows a large number of captures to be handled without mentioning them in the SDP, at the expense of not allowing the SDP to describe any of them individually. Similarly, the number of captures can vary without having to renegotiate the SDP.
(In contrast, the third TF (the teacher's face) is a separate TF because it is processed in a different role than that of the students' faces.)
In unbundled usage, there would be one transport association for the fourth TF. Incoming RTP from that association would be demultiplexed by the application based on the SSRC values, which would be unique for each student. With bundling, once the single transport TF is demultiplexed based on the ordinal index in the kumquat encapsulation, deencapsulated RTP packets destined for the fourth TF (index = 3) would be further demultiplexed by their SSRC values.
The offered SDP is:
o=- 2890844526 2890844526 IN IP4 host.example.com
c=IN IP4 10.0.1.1
a=group:KUMQUAT b c1 c2 c3 c4
The audio channel is send/receive.
m=audio 10002 RTP/AVP 0 8 97
a=mid:c1
a=label:speaker-audio
a=content:speaker
a=rtcp-mux
a=rtpmap:0 PCMU/8000
a=rtpmap:8 PCMA/8000
a=rtpmap:97 iLBC/8000
The teacher's face and presentation are send-only.
m=video 10004 RTP/AVP 103 104
a=mid:c2
a=label:speaker-video
a=content:speaker
a=sendonly
a=rtcp-mux
a=rtpmap:103 H261/90000
a=rtpmap:104 MPV/90000
m=video 10006 RTP/AVP 105 106
a=mid:c3
a=label:presentation
a=content:slides
a=sendonly
a=rtcp-mux
a=rtpmap:105 H261/90000
a=rtpmap:106 MPV/90000
The student video input is receive-only and is limited to 24
simultaneous SSRCs.
m=video 10008 RTP/AVP 105 106
a=mid:c4
a=label:student-thumbnails
a=recvonly
a=max-recv-ssrc:* 24
a=rtcp-mux
a=rtpmap:105 H261/90000
a=rtpmap:106 MPV/90000
m=multipart 10000 RTP/AVP
a=mid:b
a=rtcp-mux
a=candidate:0 1 UDP 2113601791 10.0.1.1 10000 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.32 51000 typ srflx \
raddr 10.0.1.1 rport 10000
This example contains one audio MD and two SCTP MDs, which are used for Webrtc datachannels.
o=- 2890844526 2890844526 IN IP4 host.example.com
c=IN IP4 10.0.1.1
The following group attribute declares which MDs are included in the
multiplexed MD: mid:con1 and mid:con2 are the constituent MDs whose
TFs (from the application point of view) will be carried by the TF of
the first-designed MD, mid:bundle, which is the bundle MD.
a=group:KUMQUAT bundle con1 con2
This MD provides the application-level description of the audio TF.
As in the previous example, it is the first audio m= line. It
includes any attributes which apply to the audio media from the
application point of view, including the payload type definitions.
When interpreted by a supporting processor, the transport information
is ignored. When interpreted by a processor that does not support
bundling, the transport information sets up the transport association
for the audio TF.
m=audio 10002 RTP/AVP 0 8 97
a=mid:con1
a=rtcp-mux
a=rtpmap:0 PCMU/8000
a=rtpmap:8 PCMA/8000
a=rtpmap:97 iLBC/8000
These MDs provides the the SCTP TFs. Using the Kumquat encapsulation,
the two SCTP TFs can use the same (nominal) SCTP port, since the
encapsulation carries the ordinal number of the constituent MD for
each packet. However, in this example, the two TFs use different port
numbers.
m=application 10004 DTLS/SCTP 5000
a=sctpmap:5000 webrtc-datachannel 16
m=application 10006 DTLS/SCTP 5001
a=sctpmap:5001 webrtc-datachannel 16
m=audio 10000 RTP/AVP 127
a=mid:bundle
a=rtcp-mux
a=rtpmap:127 kumquat
a=candidate:0 1 UDP 2113601791 10.0.1.1 10000 typ host
a=candidate:1 1 UDP 1694194431 198.51.100.32 51000 typ srflx \
raddr 10.0.1.1 rport 10000
TBD (Here lies the real description.)
TBD
TBD
TBD
SDP bundling uses a payload type named "kumquat" to encapsulate the RTP packets of several constituent TFs into RTP packets of one TF. Each constituent TF has a distinct index value in the range 0 to 254 (inclusive). When kumquat is used within SDP bundling, the index value is the ordinal index of the MD within the session description. (The indexes start with 0 for the first MD.)
When the application delivers a payload (and associated descriptive information such as SSRC) in the context of a constituent MD to be transmitted, it is encapsulated into a kumquat payload and the kumquat payload is transmitted using the transport association of the bundle MD.
When a kumquat payload arrives on the transport association of the bundle MD, the kumquat payload is interpreted to construct a payload (and associated descriptive information). That payload is delivered to the application in the context of the constituent MD identified by the index value.
The format of a kumquat protocol payload contains a four-octet fixed part followed by zero or more CSRC identifiers, header extension, and the encapsulated payload. Note that this diagram is of the kumquat payload only, and does not include the RTP header before the payload.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|V=2|0|X| CC |M| PT | index |transport flow |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| contributing source (CSRC) identifiers |
| .... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| extension |
| .... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| encapsulated payload |
| .... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
There is no defined meaning for the RTP marker bit in association with a kumquat payload. (Note that this is the M field of the RTP header that precedes the kumquat payload, not the M field of the kumquat payload itself.) Its value MUST be 0.
The kumquat payload represents an RTP packet containing the following data:
Graphically, the kumquat encoding sets up the following equivalence between an RTP packet of the constituent TF and an RTP packet of the bundle TF:
RTP packet in the context of the bundle media description (with PT1
specifying kumquat encoding):
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
RTP header:
+-+-+-+--+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|V=2|P|X1| 0 |0| PT1 | sequence number |
+-+-+-+--+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| timestamp |
+-+-+-+--+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| synchronization source (SSRC) identifier |
+-+-+-+--+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| extension (per X1 bit) |
| .... |
+=+=+=+==+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+
Payload of kumquat payload type:
+=+=+=+==+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+
|V=2|0|X2| CC |M| PT2 | index | 0 |
+-+-+-+--+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| extension (per X2 bit) |
| .... |
+-+-+-+--+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| contributing source (CSRC) identifiers (per CC) |
| .... |
+-+-+-+--+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| encapsulated payload |
| .... |
+-+-+-+--+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
RTP packet in the context of the constituent media description
identified by index:
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
RTP header:
+-+-+-+--+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|V=2|P|X2| CC |M| PT2 | sequence number |
+-+-+-+--+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| timestamp |
+-+-+-+--+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| synchronization source (SSRC) identifier |
+-+-+-+--+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| extension (per X2 bit) |
| .... |
+-+-+-+--+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| contributing source (CSRC) identifiers (per CC) |
| .... |
+=+=+=+==+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+
Payload of PT2 payload type:
+=+=+=+==+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+
| encapsulated payload |
| .... |
+-+-+-+--+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
The kumquat encapsulation usually adds four octets to the length of the encapsulated RTP packet. The encapsulation overhead can be larger if there is a need for a separate RTP header extension for the kumquat RTP packet.
TBD
TBD
TBD Need to discuss here how the encryption associations are set up. For SRTP/SRTCP, it would be possible to have either one association for all multiplexed streams, or one for each constituent MD, because SRTP preserves the PTs. (Have to verify that and check whether SRTCP preserves SSRCs.) But SCTP-over-DTLS can't be demultiplexed before it's decrypted, so there can be only one DTLS crypto association.
TBD
TBD
Key:
x = feature present in proposal
- = feature not present in proposal
. = feature not discussed in proposal
N/A = feature is not relevant because of another feature choice
worley-sdp-bundle-05 (KUMQUAT)
| ietf-mmusic-sdp-bundle-negotiation-03 (BUNDLE)
| | holmberg-mmusic-sdp-mmt-negotiation-00 (MMT)
| | | alvestrand-one-rtp-02 (TOGETHER)
| | | | ejzak-mmusic-bundle-alternatives-01
| | | | | Roach alternative 1a
| | | | | |(roach-mmusic-mlines-00)
| | | | | | Roach alternative 1b
| | | | | | | Roach alternative 2
| | | | | | | | westerlund-avtcore-
| | | | | | | | |transport-multiplexing-05
| | | | | | | | |(SHIM)
V V V V V V V V V
MD grouping:
one - - - - - - x - -
per type - - - - - x - - -
none x x x x x - - x x
Separate bundle MD:
no - x - x x x x x x
m=anymedia - - x - - - - - -
m=audio x - - - - - - - -
m=multipart - - - - - - - - -
Immediate update:
none - - x x x x x x x
for support x x - - - - - - -
for compat. x - - - - - - - -
Constituent MD ports after establishment:
N/A - - - - - x x - -
same - x - x - - - x x
different - - - - - - - - -
null x - - - - - - - -
rejected - - x - x - - - -
Bundle MD payload types:
N/A - - - - - x x - -
one MD - x - x . - - x x
all MDs - - x - . - - - -
encap. x - - - . - - - -
Constituent MD payload types:
N/A - - - - - x x - -
overlapping x - - - . - - x x
distinct - x x x . - - - -
Demultiplexing based on:
N/A - - - - - x x - -
PT - x x x . - - x -
encap. x - - - . - - - x
Rejection of bundle MD based on:
N/A - x - x x x x x x
media type - - x - - - - - -
proto - - - - - - - - -
codec x - - - - - - - -
Addresses/ports in constituent MDs in offer:
N/A - - - - - x x - -
NA,ZP - - - - - - - - -
NA,NZP - - - - x - - - -
RA,ZP - - - - - - - - -
RA,NZP,U x x x - x - - - -
RA,NZP,S - x - x - - - x x
NA = null address, RA = real address
ZP = zero port, NZP = non-zero port
U = unique ports, S = shared port
Are the constituent media descriptions combined into grouped media descriptions?
This proposal does not aggregate constituent MDs so that attributes can be provided directly for each constituent MD.
Is there a separate bundle media description, and if so, what media type does it have?
This proposal has a separate bundle MD so that attributes can be provided for the bundle MD independently of any constituent MD.
Is an immediate updated offer/answer used during session establishment?
This proposal requires updates for bundled answers (to tell intermediate entities to not expect media for constituent transport associations) and for non-bundled answers (to provide TURN ICE candidates, if needed).
What are the effective port numbers in MDs after the session is established?
This proposal signals null connection addresses for constituent MDs to prevent intermediate entities from expecting to see media for the constituent transport associations.
What payload types are listed for the bundled MD?
This proposal uses one payload type in the bundled media description because (1) only the encapsulated PT is sent on that transport association, and (2) to ensure that answerers that do not implement bundling reject this MD.
What is the relationship between the payload types of the constituent MDs?
This proposal allows constituent MDs to use overlapping payload types because it relies on an encapsulation to demultiplex the constituent MDs.
What is the basis for the demultiplexing of RTP?
This proposal does demultiplexing based on information in the encapsulated payload format.
We must ensure that the bundle MD is rejected by non-supporting endpoints. What method is used to ensure rejection?
This proposal uses one encapsulated payload type in the bundled media description to ensure that answerers that do not implement bundling reject this MD.
There are a number of alternative ways that the offerer can configure the constituent media descriptions.
| Method | 1 | 2 | 3 | 4 | 5 | 6 |
|---|---|---|---|---|---|---|
| Coded in chart as | NA,ZP | NA,NZP | RA,ZP | RA,NZP,U | RA,NZP,U | RA,NZP,S |
| Offered address | null | null | real | real | real | real |
| Offered port | zero | non-zero | zero | non-zero, unique | non-zero, unique | non-zero, shared |
| TURN candidates? | no | no | no | no | yes | yes |
| Supporting answerer accepts? | yes | yes | yes | yes | yes | yes |
| Update needed for supporting answerer? | no | no | possibly | yes | yes | no |
| Non-supporting answerer accepts? | no | probably | no | yes | yes | yes |
| Update needed for non-supporting answerer? | yes | yes | yes | yes | no | no |
| Disadvantages | ACE | CD | ABCE | BC | BG | BFG |
In my estimation, the worst disadvantages are A (zero port in offer), E (acceptance of offer with zero port), and F (duplicate port numbers), because they involve violations of [RFC4566] or are known to trigger limitations of large numbers of intermediate devices. Disadvantage D (offering a MD with a null address) is nearly as severe, as we expect it to cause undesired behavior in many non-supporting answerers. Disadvantages C (update needed to communicate with non-supporting answerer) and G (TURN relay must be preallocated) are moderate, and disadvantage B (updated needed to prevent intermediaries from timing out) is the least severe (because it never delays the establishment of communication).
Applying these priorities to the possible solutions, methods 4 and 5 (offer real address, non-zero unique port, with/without TURN candidates) are tied for the best choices, with the choice made based on the relative importance of minimizing preallocation of TURN relays compared to quickly establishing communication with non-supporting answerers.
This section discusses the constraints regarding demultiplexing datagrams from multiple protocols that are presented on one transport flow. This is an expansion of the analysis in [RFC5764] section 5.1.2.
The first octets of datagrams generated by particular protocols are:
| Protocol | First octet | Second octet | Third octet | Fourth octet |
|---|---|---|---|---|
| STUN | 0x00, 0x01 | 0x00, 0x01 | ||
| RTP | 0x80 to 0xBF | 0x00 to 0xC7, 0xCD to 0xFF | ||
| RTCP | 0x80 to 0xBF | 0xC8 to 0xCC | ||
| RTP/RTCPv3 | 0xC0 to 0xFF | |||
| DTLS | 0x14 to 0x17 | 0x03 | 0x03 | |
| SCTP | source port high | source port low | dest. port high | dest. port low |
TBD RFC 5764 specifies that the first octet of a DTLS packet is in the range 0x14 to 0x3F. RFC 5246 and RFC 6374 together specify the first octet is a "ContentType", with the range 0x14 to 0x17. Are additional octet values reserved for expansion? What is the range that should be reserved in practice?
The most generalized stack of protocols we consider is this:
Layer 6: ... application interfaces ...
||| ||| ||| |||
V V V V
| | | |
| | | |
Layer 5: | | | SCTP
| | | |
| | | |
RTP SRTP | |
Layer 4: RTCP SRTCP SCTP DTLS [ STUN ]
\ | | | /
--------------- ----------------
V
|
|
Layer 3: [ encapsulation STUN ]
[ \ / ]
[ ---- --------- ]
[ V ]
|
|
Layer 2: [ DTLS STUN ]
[ \ / ]
[ ------- --------- ]
[ V ]
|
|
Layer 1: [ TURN ]
|
|
Layer 0: UDP
If an SBC wishes to prevent positively the transport of certain media types or codecs, and enforces that by examining the content of RTP packets, the use of kumquat encoding may defeat the examination.
TBD
TBD
Many people have provided input for this proposal regarding both the technical aspects and the organization of the presentation. Chief among them are the authors of the predecessor proposals ([I-D.alvestrand-one-rtp] (TOGETHER), [I-D.holmberg-mmusic-sdp-mmt-negotiation] (MMT), and [I-D.ietf-mmusic-sdp-bundle-negotiation] (BUNDLE)): Harald Alvestrand, Jonathan Lennox, and Christer Holmberg. In addition, input was provided by Charles Eckel, Andrew Hutton, Cullen Jennings, Hadriel Kaplan, Paul Kyzivat, Adam Roach, and Robert Sparks.
Note to RFC Editor: Please remove this section before publication.
Initial version.
Thoroughly revise the text and structure of the document.
Heavily revise Terminology regarding media flows.
Revise Desiderata, including adding that multiple separate bundles must be possible, and noninterference with ICE negotiation.
Add section on ICE considerations.
Change "fusion" to "bundle".
Use a=rtcp-mux in examples to be more realistic (and to shorten the examples).
Correct the use of ICE in answers; ICE candidates are not provided if an offered MD does not contain ICE candidates.
Add an example of a fast-start offer.
Add design comparison Section 7.
Use bibxml references.
Add DES C9, regarding continued usage of transcoding facilities offered by intermediate entities.
Add demultiplexing considerations Section 8.
Change recommendation for SCTP port numbers from 0xC000-0xFFFF to 0x4000-0x7FFF to avoid collision with a future RTP/RTCP version 3.
Add the transport flow index to the KUMQUAT encapsulation.
Add section on choices for offering constituent MDs Section 7.9. Revise the examples to show offering "real address, non-zero port, no ICE candidates".
Add note to DES C9 (support intermediate transcoding facilities) saying that intermediate transcoding facilities are not expected to be very useful, given that encryption will be the normal use case.
Add an exampleSection 4.4 with SCTP MDs.
Add a section for encryption considerations.Section 5.7
Revise generalized demultiplexing diagram to make explicit the optional RTP encapsulation layer.
Update comparison chartSection 7 for draft-ejzak-mmusic-bundle-alternatives-01[I-D.ejzak-mmusic-bundle-alternatives] and draft-westerlund-avtcore-transport-multiplexing-05[I-D.westerlund-avtcore-transport-multiplexing].
Update comparison chartSection 7 to discuss alternative address/port/candidate policies for offering constituent MDs.
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997. |
| [RFC3264] | Rosenberg, J. and H. Schulzrinne, "An Offer/Answer Model with Session Description Protocol (SDP)", RFC 3264, DOI 10.17487/RFC3264, June 2002. |
| [RFC3550] | Schulzrinne, H., Casner, S., Frederick, R. and V. Jacobson, "RTP: A Transport Protocol for Real-Time Applications", STD 64, RFC 3550, DOI 10.17487/RFC3550, July 2003. |
| [RFC4566] | Handley, M., Jacobson, V. and C. Perkins, "SDP: Session Description Protocol", RFC 4566, DOI 10.17487/RFC4566, July 2006. |
| [RFC5245] | Rosenberg, J., "Interactive Connectivity Establishment (ICE): A Protocol for Network Address Translator (NAT) Traversal for Offer/Answer Protocols", RFC 5245, DOI 10.17487/RFC5245, April 2010. |
| [RFC5888] | Camarillo, G. and H. Schulzrinne, "The Session Description Protocol (SDP) Grouping Framework", RFC 5888, DOI 10.17487/RFC5888, June 2010. |
| [RFC2327] | Handley, M. and V. Jacobson, "SDP: Session Description Protocol", RFC 2327, DOI 10.17487/RFC2327, April 1998. |
| [RFC3725] | Rosenberg, J., Peterson, J., Schulzrinne, H. and G. Camarillo, "Best Current Practices for Third Party Call Control (3pcc) in the Session Initiation Protocol (SIP)", BCP 85, RFC 3725, DOI 10.17487/RFC3725, April 2004. |
| [RFC4796] | Hautakorpi, J. and G. Camarillo, "The Session Description Protocol (SDP) Content Attribute", RFC 4796, DOI 10.17487/RFC4796, February 2007. |
| [RFC4960] | Stewart, R., "Stream Control Transmission Protocol", RFC 4960, DOI 10.17487/RFC4960, September 2007. |
| [RFC5761] | Perkins, C. and M. Westerlund, "Multiplexing RTP Data and Control Packets on a Single Port", RFC 5761, DOI 10.17487/RFC5761, April 2010. |
| [RFC5764] | McGrew, D. and E. Rescorla, "Datagram Transport Layer Security (DTLS) Extension to Establish Keys for the Secure Real-time Transport Protocol (SRTP)", RFC 5764, DOI 10.17487/RFC5764, May 2010. |
| [I-D.alvestrand-mmusic-msid] | Alvestrand, H., "Cross Session Stream Identification in the Session Description Protocol", Internet-Draft draft-alvestrand-mmusic-msid-02, December 2012. |
| [I-D.alvestrand-one-rtp] | Alvestrand, H., "SDP Grouping for Single RTP Sessions", Internet-Draft draft-alvestrand-one-rtp-02, September 2011. |
| [I-D.ejzak-mmusic-bundle-alternatives] | Ejzak, R., "Alternatives to BUNDLE", Internet-Draft draft-ejzak-mmusic-bundle-alternatives-01, February 2013. |
| [I-D.holmberg-mmusic-sdp-mmt-negotiation] | Holmberg, C., Alvestrand, H. and J. Lennox, "Multiplexed Media Types (MMT) Using Session Description Protocol (SDP) Port Numbers", Internet-Draft draft-holmberg-mmusic-sdp-mmt-negotiation-00, October 2012. |
| [I-D.ietf-avtcore-multi-media-rtp-session] | Westerlund, M., Perkins, C. and J. Lennox, "Sending Multiple Types of Media in a Single RTP Session", Internet-Draft draft-ietf-avtcore-multi-media-rtp-session-13, December 2015. |
| [I-D.ietf-mmusic-sdp-bundle-negotiation] | Holmberg, C., Alvestrand, H. and C. Jennings, "Negotiating Media Multiplexing Using the Session Description Protocol (SDP)", Internet-Draft draft-ietf-mmusic-sdp-bundle-negotiation-31, June 2016. |
| [I-D.jennings-mmusic-media-req] | Jennings, C., Uberti, J. and E. Rescorla, "Requirements from various WG for MMUSIC", Internet-Draft draft-jennings-mmusic-media-req-00, February 2013. |
| [I-D.roach-mmusic-mlines] | Roach, A., "Thoughts on syntax for representing multiple media streams", Internet-Draft draft-roach-mmusic-mlines-00, January 2013. |
| [I-D.westerlund-avtcore-transport-multiplexing] | Westerlund, M. and C. Perkins, "Multiplexing Multiple RTP Sessions onto a Single Lower-Layer Transport", Internet-Draft draft-westerlund-avtcore-transport-multiplexing-07, October 2013. |
| [I-D.worley-service-example] | Worley, D., "Session Initiation Protocol Service Example -- Music on Hold", Internet-Draft draft-worley-service-example-15, November 2013. |