Internet Engineering Task Force (IETF) G. Camarillo Request for Comments: 5888 Ericsson Obsoletes: 3388 H. Schulzrinne Category: Standards Track Columbia University ISSN: 2070-1721 June 2010
The Session Description Protocol (SDP) Grouping Framework
Abstract
In this specification, we define a framework to group "m" lines in the Session Description Protocol (SDP) for different purposes. This framework uses the "group" and "mid" SDP attributes, both of which are defined in this specification. Additionally, we specify how to use the framework for two different purposes: for lip synchronization and for receiving a media flow consisting of several media streams on different transport addresses. This document obsoletes RFC 3388.
Status of This Memo
This is an Internet Standards Track document.
This document is a product of the Internet Engineering Task Force (IETF). It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG). Further information on Internet Standards is available in Section 2 of RFC 5741.
Information about the current status of this document, any errata, and how to provide feedback on it may be obtained at http://www.rfc-editor.org/info/rfc5888.
Copyright Notice
Copyright (c) 2010 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.
An SDP [RFC4566] session description typically contains one or more media lines, which are commonly known as "m" lines. When a session description contains more than one "m" line, SDP does not provide any means to express a particular relationship between two or more of them. When an application receives an SDP session description with more than one "m" line, it is up to the application to determine what to do with them. SDP does not carry any information about grouping media streams.
While in some environments this information can be carried out of band, it is necessary to have a mechanism in SDP to express how different media streams within a session description relate to each other. The framework defined in this specification is such a mechanism.
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].
This section provides a non-normative description of how the SDP Grouping Framework defined in this document works. In a given session description, each "m" line is identified by a token, which is carried in a "mid" attribute below the "m" line. The session description carries session-level "group" attributes that group different "m" lines (identified by their tokens) using different group semantics. The semantics of a group describe the purpose for which the "m" lines are grouped. For example, the "group" line in the session description below indicates that the "m" lines identified by tokens 1 and 2 (the audio and the video "m" lines, respectively) are grouped for the purpose of lip synchronization (LS).
This document defines the "media stream identification" media attribute, which is used for identifying media streams within a session description. Its formatting in SDP [RFC4566] is described by the following Augmented Backus-Naur Form (ABNF) [RFC5234]:
mid-attribute = "a=mid:" identification-tag identification-tag = token ; token is defined in RFC 4566
The identification-tag MUST be unique within an SDP session description.
This document defines the "group" session-level attribute, which is used for grouping together different media streams. Its formatting in SDP is described by the following ABNF [RFC5234]:
group-attribute = "a=group:" semantics *(SP identification-tag) semantics = "LS" / "FID" / semantics-extension semantics-extension = token ; token is defined in RFC 4566
This document defines two standard semantics: Lip Synchronization (LS) and Flow Identification (FID). Semantics extensions follow the Standards Action policy [RFC5226].
All of the "m" lines of a session description that uses "group" MUST be identified with a "mid" attribute whether they appear in the group line(s) or not. If a session description contains at least one "m" line that has no "mid" identification, the application MUST NOT perform any grouping of media lines.
Camarillo & Schulzrinne Standards Track [Page 4]
RFC 5888 SDP Grouping Framework June 2010
"a=group" lines are used to group together several "m" lines that are identified by their "mid" attribute. "a=group" lines that contain identification-tags that do not correspond to any "m" line within the session description MUST be ignored. The application acts as if the "a=group" line did not exist. The behavior of an application receiving an SDP description with grouped "m" lines is defined by the semantics field in the "a=group" line.
There MAY be several "a=group" lines in a session description. The "a=group" lines of a session description can use the same or different semantics. An "m" line identified by its "mid" attribute MAY appear in more than one "a=group" line.
An application that receives a session description that contains "m" lines that are grouped together using LS semantics MUST synchronize the playout of the corresponding media streams. Note that LS semantics apply not only to a video stream that has to be synchronized with an audio stream; the playout of two streams of the same type can be synchronized as well.
For RTP streams, synchronization is typically performed using the RTP Control Protocol (RTCP), which provides enough information to map time stamps from the different streams into a local absolute time value. However, the concept of media stream synchronization MAY also apply to media streams that do not make use of RTP. If this is the case, the application MUST recover the original timing relationship between the streams using whatever mechanism is available.
The following example shows a session description of a conference that is being multicast. The first media stream (mid:1) contains the voice of the speaker who speaks in English. The second media stream (mid:2) contains the video component, and the third (mid:3) media stream carries the translation to Spanish of what she is saying. The first and second media streams have to be synchronized.
An "m" line in an SDP session description defines a media stream. However, SDP does not define what a media stream is. This definition can be found in the Real Time Streaming Protocol (RTSP) specification. The RTSP RFC [RFC2326] defines a media stream as "a single media instance, e.g., an audio stream or a video stream as well as a single whiteboard or shared application group. When using RTP, a stream consists of all RTP and RTCP packets created by a source within an RTP session".
This definition assumes that a single audio (or video) stream maps into an RTP session. The RTP RFC [RFC1889] (at present obsoleted by [RFC3550]) used to define an RTP session as follows: "For each participant, the session is defined by a particular pair of destination transport addresses (one network address plus a port pair for RTP and RTCP)".
While the previous definitions cover the most common cases, there are situations where a single media instance (e.g., an audio stream or a video stream) is sent using more than one RTP session. Two examples (among many others) of this kind of situation are cellular systems using the Session Initiation Protocol (SIP; [RFC3261]) and systems receiving Dual-Tone Multi-Frequency (DTMF) tones on a different host than the voice.
Systems using a cellular access and SIP as a signalling protocol need to receive media over the air. During a session, the media can be encoded using different codecs. The encoded media has to traverse
Camarillo & Schulzrinne Standards Track [Page 6]
RFC 5888 SDP Grouping Framework June 2010
the radio interface. The radio interface is generally characterized as being prone to bit errors and associated with relatively high packet transfer delays. In addition, radio interface resources in a cellular environment are scarce and thus expensive, which calls for special measures in providing a highly efficient transport. In order to get an appropriate speech quality in combination with an efficient transport, precise knowledge of codec properties is required so that a proper radio bearer for the RTP session can be configured before transferring the media. These radio bearers are dedicated bearers per media type (i.e., codec).
Cellular systems typically configure different radio bearers on different port numbers. Therefore, incoming media has to have different destination port numbers for the different possible codecs in order to be routed properly to the correct radio bearer. Thus, this is an example in which several RTP sessions are used to carry a single media instance (the encoded speech from the sender).
Some voice sessions include DTMF tones. Sometimes, the voice handling is performed by a different host than the DTMF handling. It is common to have an application server in the network gathering DTMF tones for the user while the user receives the encoded speech on his user agent. In this situation, it is necessary to establish two RTP sessions: one for the voice and the other for the DTMF tones. Both RTP sessions are logically part of the same media instance.
The previous examples show that the definition of a media stream in [RFC2326] does not cover some scenarios. It cannot be assumed that a single media instance maps into a single RTP session. Therefore, we introduce the definition of a media flow:
A media flow consists of a single media instance, e.g., an audio stream or a video stream as well as a single whiteboard or shared application group. When using RTP, a media flow comprises one or more RTP sessions.
Several "m" lines grouped together using FID semantics form a media flow. A media agent handling a media flow that comprises several "m" lines MUST send a copy of the media to every "m" line that is part of the flow as long as the codecs and the direction attribute present in a particular "m" line allow it.
Camarillo & Schulzrinne Standards Track [Page 7]
RFC 5888 SDP Grouping Framework June 2010
It is assumed that the application uses only one codec at a time to encode the media produced. This codec MAY change dynamically during the session, but at any particular moment, only one codec is in use.
The application encodes the media using the current codec and checks, one by one, all of the "m" lines that are part of the flow. If a particular "m" line contains the codec being used and the direction attribute is "sendonly" or "sendrecv", a copy of the encoded media is sent to the address/port specified in that particular media stream. If either the "m" line does not contain the codec being used or the direction attribute is neither "sendonly" nor "sendrecv", nothing is sent over this media stream.
The application typically ends up sending media to different destinations (IP address/port number) depending on the codec used at any moment.
The session description below might be sent by a SIP user agent using a cellular access. The user agent supports GSM (Global System for Mobile communications) on port 30000 and AMR (Adaptive Multi-Rate) on port 30002. When the remote party sends GSM, it will send RTP packets to port number 30000. When AMR is the codec chosen, packets will be sent to port 30002. Note that the remote party can switch between both codecs dynamically in the middle of the session. However, in this example, only one media stream at a time carries voice. The other remains "muted" while its corresponding codec is not in use.
(The linebreak in the fmtp line accommodates RFC formatting restrictions; SDP does not have continuation lines.)
Camarillo & Schulzrinne Standards Track [Page 8]
RFC 5888 SDP Grouping Framework June 2010
In the previous example, a system receives media on the same IP address on different port numbers. The following example shows how a system can receive different codecs on different IP addresses.
(The linebreak in the fmtp line accommodates RFC formatting restrictions; SDP does not have continuation lines.)
The cellular terminal in this example only supports the AMR codec. However, many current IP phones only support PCM (Pulse-Code Modulation; payload 0). In order to be able to interoperate with them, the cellular terminal uses a transcoder whose IP address is 192.0.2.2. The cellular terminal includes the transcoder IP address in its SDP description to provide support for PCM. Remote systems will send AMR directly to the terminal, but PCM will be sent to the transcoder. The transcoder will be configured (using whatever method is preferred) to convert the incoming PCM audio to AMR and send it to the terminal.
The next example shows how the "group" attribute used with FID semantics can indicate the use of two different codecs in the two directions of a bidirectional media stream.
A user agent that receives the SDP description above knows that, at a certain moment, it can send either PCM u-law to port number 30000 or PCM A-law to port number 30002. However, the media agent also knows that the other end will only send PCM u-law (payload 0).
The following example shows a session description with different "m" lines grouped together using FID semantics that contain the same codec.
At a particular point in time, if the media agent receiving the SDP message above is sending PCM u-law (payload 0), it sends RTP packets to 192.0.2.1 on port 30000 and to 192.0.2.2 on port 20000 (first and third "m" lines). If it is sending PCM A-law (payload 8), it sends RTP packets to 192.0.2.1 on port 30002 and to 192.0.2.2 on port 20000 (second and third "m" lines).
The system that generated the SDP description above supports PCM u-law on port 30000 and PCM A-law on port 30002. Besides, it uses an application server that records the conversation and whose IP address is 192.0.2.2. The application server does not need to understand the media content, so it always receives a copy of the media stream, regardless of the codec and payload type that is being used. That is why the application server always receives a copy of the audio stream regardless of the codec being used at any given moment (it actually performs an RTP dump, so it can effectively receive any codec).
Remember that if several "m" lines that are grouped together using the FID semantics contain the same codec, the media agent MUST send copies of the same media stream as several RTP sessions at the same time.
The last example in this section deals with DTMF tones. DTMF tones can be transmitted using a regular voice codec or can be transmitted as telephony events. The RTP payload for DTMF tones treated as
Camarillo & Schulzrinne Standards Track [Page 10]
RFC 5888 SDP Grouping Framework June 2010
telephone events is described in [RFC4733]. Below, there is an example of an SDP session description using FID semantics and this payload type.
The remote party would send PCM encoded voice (payload 0) to 192.0.2.1 and DTMF tones encoded as telephony events to 192.0.2.2. Note that only voice or DTMF is sent at a particular point in time. When DTMF tones are sent, the first media stream does not carry any data and, when voice is sent, there is no data in the second media stream. FID semantics provide different destinations for alternative codecs.
It is worthwhile mentioning some scenarios where the "group" attribute using existing semantics (particularly FID) might seem to be applicable but is not.
FID semantics are useful when the application only uses one codec at a time. An application that encodes the same media using different codecs simultaneously MUST NOT use FID to group those media lines. Some systems that handle DTMF tones are a typical example of parallel encoding using different codecs. Some systems implement the RTP payload defined in RFC 4733 [RFC4733], but when they send DTMF tones, they do not mute the voice channel. Therefore, in effect they are sending two copies of the same DTMF tone: encoded as voice and encoded as a telephony event. When the receiver gets both copies, it typically uses the telephony event rather than the tone encoded as voice. FID semantics MUST NOT be used in this context to group both media streams, since such a system is not using alternative codecs but rather different parallel encodings for the same information.
Layered encoding schemes encode media in different layers. The quality of the media stream at the receiver varies depending on the number of layers received. SDP provides a means to group together contiguous multicast addresses that transport different layers. The "c" line below:
FID MUST NOT be used to group "m" lines that do not represent the same information. Therefore, FID MUST NOT be used to group "m" lines that contain the different layers of layered encoding schemes. Besides, we do not define new group semantics to provide a more flexible way of grouping different layers, because the already existing SDP mechanism covers the most useful scenarios. Since the existing SDP mechanism already covers the most useful scenarios, we do not define a new group semantics to define a more flexible way of grouping different layers.
If media streams using several different codecs have to be sent to the same IP address and port, the traditional SDP syntax of listing several codecs in the same "m" line MUST be used. FID MUST NOT be used to group "m" lines with the same IP address/port. Therefore, an SDP description like the one below MUST NOT be generated.
SDP descriptions are used by several different protocols, SIP among them. We include a section about SIP, because the "group" attribute will most likely be used mainly by SIP systems.
SIP [RFC3261] is an application layer protocol for establishing, terminating, and modifying multimedia sessions. SIP carries session descriptions in the bodies of the SIP messages but is independent from the protocol used for describing sessions. SDP [RFC4566] is one of the protocols that can be used for this purpose.
At session establishment, SIP provides a three-way handshake (INVITE-200 OK-ACK) between end systems. However, just two of these three messages carry SDP, as described in [RFC3264].
The "mid" attribute is an identifier for a particular media stream. Therefore, the "mid" value in the offer MUST be the same as the "mid" value in the answer. Besides, subsequent offers (e.g., in a re-INVITE) SHOULD use the same "mid" value for the already existing media streams.
[RFC3264] describes the usage of SDP in text of SIP. The offerer and the answerer align their media description so that the nth media stream ("m=" line) in the offerer's session description corresponds to the nth media stream in the answerer's description.
The presence of the "group" attribute in an SDP session description does not modify this behavior.
Since the "mid" attribute provides a means to label "m" lines, it would be possible to perform media alignment using "mid" labels rather than matching nth "m" lines. However, this would not bring any gain and would add complexity to implementations. Therefore, SIP
Camarillo & Schulzrinne Standards Track [Page 13]
RFC 5888 SDP Grouping Framework June 2010
systems MUST perform media alignment matching nth lines regardless of the presence of the "group" or "mid" attributes.
If a media stream that contained a particular "mid" identifier in the offer contains a different identifier in the answer, the application ignores all of the "mid" and "group" lines that might appear in the session description. The following example illustrates this scenario.
Since alignment of "m" lines is performed based on matching of nth lines, the first stream had "mid:1" in the INVITE and "mid:2" in the 200 OK. Therefore, the application ignores every "mid" and "group" line contained in the SDP description.
Camarillo & Schulzrinne Standards Track [Page 14]
RFC 5888 SDP Grouping Framework June 2010
A well-behaved SIP user agent would have returned the SDP description below in the 200 OK response.
A SIP entity that receives an offer that contains an "a=group" line with semantics that it does not understand MUST return an answer without the "group" line. Note that, as described in the previous section, the "mid" lines MUST still be present in the answer.
A SIP entity that receives an offer that contains an "a=group" line with semantics that are understood MUST return an answer that contains an "a=group" line with the same semantics. The identification-tags contained in this "a=group" line MUST be the same as those received in the offer, or a subset of them (zero identification-tags is a valid subset). When the identification-tags in the answer are a subset, the "group" value to be used in the session MUST be the one present in the answer.
SIP entities refuse media streams by setting the port to zero in the corresponding "m" line. "a=group" lines MUST NOT contain identification-tags that correspond to "m" lines with the port set to zero.
Note that grouping of "m" lines MUST always be requested by the offerer, but never by the answerer. Since SIP provides a two-way SDP exchange, an answerer that requested grouping would not know whether the "group" attribute was accepted by the offerer or not. An answerer that wants to group media lines issues another offer after having responded to the first one (in a re-INVITE, for instance).
The example below shows how the callee refuses a media stream offered by the caller by setting its port number to zero. The "mid" value corresponding to that media stream is removed from the "group" value in the answer.
Camarillo & Schulzrinne Standards Track [Page 15]
RFC 5888 SDP Grouping Framework June 2010
SDP description in the INVITE from caller to callee:
A client that understands "group" and "mid", but does not want to use these SDP features in a particular session, may still want to indicate that it supports these features. To indicate this support, a client can add an "a=3Dgroup" line with no identification-tags for every semantics value it understands.
If a server receives an offer that contains empty "a=group" lines, it SHOULD add its capabilities also in the form of empty "a=group" lines to its answer.
A system that supports both LS and FID semantics but does not want to group any media stream for this particular session generates the following SDP description:
This document does not define any SIP "Require" header field. Therefore, if one of the SIP user agents does not understand the "group" attribute, the standard SDP fall-back mechanism MUST be used, namely, attributes that are not understood are simply ignored.
This situation does not represent a problem, because grouping requests are always performed by offerers and not by answerers. If the offerer does not support "group", this attribute will simply not be used.
The answerer will ignore the "group" attribute since it does not understand it and will also ignore the "mid" attribute. For LS semantics, the answerer might decide to perform, or not to perform, synchronization between media streams.
For FID semantics, the answerer will consider the session to consist of several media streams.
Different implementations will behave in different ways.
In the case of audio and different "m" lines for different codecs, an implementation might decide to act as a mixer with the different incoming RTP sessions, which is the correct behavior.
Camarillo & Schulzrinne Standards Track [Page 17]
RFC 5888 SDP Grouping Framework June 2010
An implementation might also decide to refuse the request (e.g., 488 Not Acceptable Here, or 606 Not Acceptable), because it contains several "m" lines. In this case, the server does not support the type of session that the caller wanted to establish. In case the client is willing to establish a simpler session anyway, the client can re-try the request without the "group" attribute and with only one "m" line per flow.
Section 3 (Overview of Operation) has been added for clarity. The AMR and GSM acronyms are now expanded on their first use. The examples now use IP addresses in the range suitable for examples.
The grouping mechanism is now defined as an extensible framework. Earlier, RFC 3388 [RFC3388] used to discourage extensions to this mechanism in favor of using new session description protocols.
Given a semantics value, RFC 3388 [RFC3388] used to restrict "m" line identifiers to only appear in a single group using that semantics. That restriction has been lifted in this specification. From conversations with implementers, existing (i.e., legacy) implementations enforce this restriction on a per-semantics basis. That is, they only enforce this restriction for supported semantics. Because of the nature of existing semantics, implementations will only use a single "m" line identifier across groups using a given semantics even after the restriction has been lifted by this specification. Consequently, the lifting of this restriction will not cause backward-compatibility problems, because implementations supporting new semantics will be updated to not enforce this restriction at the same time as they are updated to support the new semantics.
Using the "group" parameter with FID semantics, an entity that managed to modify the session descriptions exchanged between the participants to establish a multimedia session could force the participants to send a copy of the media to any destination of its choosing.
Integrity mechanisms provided by protocols used to exchange session descriptions and media encryption can be used to prevent this attack. In SIP, Secure/Multipurpose Internet Mail Extensions (S/MIME) [RFC5750] and Transport Layer Security (TLS) [RFC5246] can be used to protect session description exchanges in an end-to-end and a hop-by- hop fashion, respectively.
This document defines two SDP attributes: "mid" and "group".
The "mid" attribute is used to identify media streams within a session description, and its format is defined in Section 4.
The "group" attribute is used for grouping together different media streams, and its format is defined in Section 5.
This document defines a framework to group media lines in SDP using different semantics. Semantics values to be used with this framework are registered by the IANA following the Standards Action policy [RFC5226].
The IANA Considerations section of the RFC MUST include the following information, which appears in the IANA registry along with the RFC number of the publication.
o A brief description of the semantics.
o Token to be used within the "group" attribute. This token may be of any length, but SHOULD be no more than four characters long.
o Reference to a standards track RFC.
The following are the current entries in the registry:
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997.
[RFC3261] Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston, A., Peterson, J., Sparks, R., Handley, M., and E. Schooler, "SIP: Session Initiation Protocol", RFC 3261, June 2002.
[RFC3264] Rosenberg, J. and H. Schulzrinne, "An Offer/Answer Model with Session Description Protocol (SDP)", RFC 3264, June 2002.
[RFC4566] Handley, M., Jacobson, V., and C. Perkins, "SDP: Session Description Protocol", RFC 4566, July 2006.
[RFC5226] Narten, T. and H. Alvestrand, "Guidelines for Writing an IANA Considerations Section in RFCs", BCP 26, RFC 5226, May 2008.
[RFC5234] Crocker, D. and P. Overell, "Augmented BNF for Syntax Specifications: ABNF", STD 68, RFC 5234, January 2008.
[RFC5246] Dierks, T. and E. Rescorla, "The Transport Layer Security (TLS) Protocol Version 1.2", RFC 5246, August 2008.
[RFC5750] Ramsdell, B. and S. Turner, "Secure/Multipurpose Internet Mail Extensions (S/MIME) Version 3.2 Certificate Handling", RFC 5750, January 2010.
[RFC1889] Schulzrinne, H., Casner, S., Frederick, R., and V. Jacobson, "RTP: A Transport Protocol for Real-Time Applications", RFC 1889, January 1996.
[RFC2326] Schulzrinne, H., Rao, A., and R. Lanphier, "Real Time Streaming Protocol (RTSP)", RFC 2326, April 1998.
[RFC3388] Camarillo, G., Eriksson, G., Holler, J., and H. Schulzrinne, "Grouping of Media Lines in the Session Description Protocol (SDP)", RFC 3388, December 2002.
Camarillo & Schulzrinne Standards Track [Page 20]
RFC 5888 SDP Grouping Framework June 2010
[RFC3550] Schulzrinne, H., Casner, S., Frederick, R., and V. Jacobson, "RTP: A Transport Protocol for Real-Time Applications", STD 64, RFC 3550, July 2003.
[RFC4733] Schulzrinne, H. and T. Taylor, "RTP Payload for DTMF Digits, Telephony Tones, and Telephony Signals", RFC 4733, December 2006.
Authors' Addresses
Gonzalo Camarillo Ericsson Hirsalantie 11 Jorvas 02420 FINLAND
EMail: Gonzalo.Camarillo@ericsson.com
Henning Schulzrinne Columbia University 1214 Amsterdam Avenue New York, NY 10027 USA