]> Vidyo, Inc.

433 Hackensack Avenue Sixth Floor Hackensack NJ 07601 US jonathan@vidyo.com

Helsinki University of Technology (TKK)

Networking Laboratory PO Box 3000 FIN-02015 TKK Finland jo@acm.org

Fraunhofer HHI

Einsteinufer 37 D-10587 Berlin Germany +49-30-31002-227 schierl@hhi.fhg.de

RAI MMUSIC I-D Internet-Draft The Session Description Protocol provides mechanisms to describe attributes of multimedia sessions and of individual media streams (e.g., Real-time Transport Protocol (RTP) sessions) within a multimedia session, but does not provide any mechanism to describe individual media sources within a media stream. This document defines a mechanism to describe RTP media sources, identified by their Synchronization Source Identifiers (SSRCs), in SDP, associate attributes with these sources, and express relationships among sources. It also defines several source-level attributes which can be used to describe properties of media sources.

The Session Description Protocol (SDP) provides mechanisms to describe attributes of multimedia sessions and of media streams (e.g., Real-time Transport Protocol (RTP) sessions) within a multimedia session, but does not provide any mechanism to describe individual media sources within a media stream. Several recently-proposed protocols, notably RTP Single-Source Multicast have found it useful to describe specific media sources in SDP messages. Single-source multicast, in particular, needs to ensure that receivers' RTP Synchronization Source Identifiers (SSRCs) do not collide with those of media senders, as the RTP specification requires that colliding sources change their SSRC values after a collision has been detected. Earlier work has used mechanisms specific to each protocol to describe the individual sources of an RTP session. Moreover, whereas the Real-Time Transport Protocol (RTP) is defined as allowing multiple sources in an RTP session (for example, if a user has more than one camera), SDP has no existing mechanism for an endpoint to indicate that it will be using multiple sources, or to describe their characteristics individually. To address all these problems, this document defines a mechanism to describe RTP sources, identified by their Synchronization Sources Identifiers (SSRCs), in SDP, associate attributes with these sources, and express relationships among individual sources. It also defines a number of new SDP attributes that apply to individual sources ("source-level" attributes); describes how a number of existing media stream ("media-level") attributes can also be applied at the source level; and establishes IANA registries for source-level attributes and source grouping semantics.

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 RFC 2119 and indicate requirement levels for compliant implementations.

In the Real-Time Transport Protocol (RTP), an association among a group of communicating participants is known as an RTP Session. An RTP session is typically associated with a single transport address (in the case of multicast) or communication flow (in the case of unicast), though RTP translators and single-source multicast can make the situation more complex. RTP topologies are discussed in more detail in . Within an RTP session, the source of a single stream of RTP packets is known as a synchronization source (SSRC). Every synchronization source is identified by a 32-bit numeric identifier. In addition, receivers (who may never send RTP packets) also have source identifiers, which are used to identify their RTP Control Protocol (RTCP) receiver reports and other feedback messages. Messages of the Session Description Protocol (SDP), known as Session Descriptions, describe Multimedia Sessions. A multimedia session is a set of multimedia senders and receivers, and the data streams flowing from senders to receivers. A multimedia session contains a number of Media Streams, which are the individual RTP sessions or other media paths over which one type of multimedia data is carried. Information that applies to an entire multimedia session is called Session-Level information, while information pertaining to one media stream is called Media-Level information. The collection of all the information describing a media stream is known as a Media Description. (Media descriptions are also sometimes known informally as SDP "m"-lines, after the SDP syntax that begins a media description.) Several standard information elements are defined at both the session level and the media level. Extended information can be included at both levels through the use of attributes. (The term "Media Stream" does not appear in the SDP specification itself, but is used by a number of SDP extensions, for instance Interactive Connectivity Establishment (ICE), to denote the object described by an SDP media description. This term is unfortunately rather confusing, as the RTP specification uses the term "media stream" to refer to an individual media source or RTP packet stream, identified by an SSRC, whereas an SDP media stream describes an entire RTP session, which can contain any number of RTP sources. In this document, the term "media stream" means an SDP media stream, i.e. the thing described by an SDP media description, whereas "media source" is used for a single source of media packets, i.e. an RTP media stream.) The core SDP specification does not have any way of describing individual media sources, in particular RTP synchronization sources, within a media stream. To address this problem, in this document we introduce a third level of information, called Source-Level information. Syntactically, source-level information is described by a new SDP media-level attribute "ssrc", which identifies specific synchronization sources within an RTP session, and acts as a meta-attribute mapping source-level attribute information to these sources. This document also defines an SDP media-level attribute "ssrc-group", which can represent relationships among media sources within an RTP session, in much the same way as the "group" attribute represents relationships among media streams within a multimedia session.

This section defines two media-level attributes, "ssrc" and "ssrc-group".

a=ssrc: : ]]>

The SDP media attribute "ssrc" indicates a property (known as a "source-level attribute") of a media source (RTP stream) within an RTP session. ]]> is the synchronizaton source ID (SSRC) of the source being described, interpreted as a 32-bit unsigned integer in network byte order and represented in decimal. ]]> or :]]> represent the source-level attribute specific to the given media source. The source-level attribute follows the syntax of the SDP "a=" line. It thus consists either of a single attribute name (a flag), or an attribute name and value, e.g. "cname:user@example.com". No attributes of the former type are defined by this document. Within a media stream, ssrc attributes with the same value of ]]> describe different attributes of the same media sources. Across media streams, ]]> values are not correlated (unless correlation is indicated by media-stream grouping or some other mechanism) and MAY be repeated. Each "ssrc" media attribute specifies a single source-level attribute for the given ]]>. For each source mentioned in SDP, the source-level attribute "cname", defined in , MUST be provided. Any number of other source-level attributes for the source MAY also be provided. The "ssrc" media attribute MAY be used for any RTP-based media transport. It is not defined for other transports. If any other SDP attributes also mention RTP SSRC values (for example, MIKEY), the values used MUST be consistent. (These attributes MAY provide additional information about a source described by an "ssrc" attribute, or MAY describe additional sources.) Though the source-level attributes specified by the ssrc property follow the same syntax as session-level and media-level attributes, they are defined independently. All source-level attributes MUST be registered with IANA, using the registry defined in . in gives a formal Augmented Backus-Naur Form (ABNF) grammar for the ssrc attribute. The "ssrc" media attribute is not dependent on charset.

... ]]>

The SDP media attribute "ssrc-group" expresses a relationship among several sources of an RTP session. It is analogous to the "group" session-level attribute, which expresses a relationship among media streams in an SDP multimedia session (i.e., a relationship among several logically related RTP sessions). As sources are already identified by their SSRC IDs, no analogous property to the "mid" attribute is necessary; groups of sources are identified by their SSRC IDs directly. The ]]> parameter is taken from the specification of the "group" attribute. The initial semantics values defined for the ssrc-group attribute are FID (Flow Identification) and FEC (Forward Error Correction). In each case, the relationship among the grouped sources is the same as the relationship among corresponding sources in media streams grouped using the SDP "group" attribute. Though the "ssrc-group" semantics values follow the same syntax as "group" semantics values, they are defined independently. All "ssrc-group" semantics values MUST be registered with IANA, using the registry defined in . (The other "group" semantics registered with IANA as of this writing are not useful for source grouping. LS (Lip Synchronization) is redundant for sources within a media stream, as RTP sources with the same CNAME are implicitly synchronized in RTP. SRF (Single Reservation Flow) and ANAT (Alternative Network Address Types) refer specifically to the media stream's transport characteristics. CS (Composite Session) is used to group FLUTE sessions, and so is not applicable to RTP.) The ssrc-group attribute indicates the sources in a group by listing the ]]>s of the sources in the group. It MUST list at least one ]]> for a group, and MAY list any number of additional ones. Every ]]> listed in an ssrc-group attribute MUST be defined by a corresponding "ssrc:" line in the same media description. The "ssrc-group" media attribute is not dependent on charset. in gives a formal Augmented Backus-Naur Form (ABNF) grammar for the ssrc-group attribute.

The synchronization source identifiers used in an RTP session are chosen randomly and independently by endpoints. As such, it is possible for two RTP endpoints to choose the same SSRC identifier. Though the probability of this is low, the RTP specification requires that all RTP endpoints MUST be prepared to detect and resolve collisions. As a result, all endpoints MUST be prepared for the fact that information about specific sources identified in a media stream might be out of date. The actual binding between SSRCs and source CNAMEs can only be identified by the source description (SDES) RTCP packets transmitted on the RTP session. When endpoints are choosing their own local SSRC values for media streams for which source-level attributes have been specified, they MUST NOT use for themselves any SSRC identifiers mentioned in media descriptions they have received for the media stream. However, sources identified by SDP source-level attributes do not otherwise affect RTP transport logic. Specifically, sources which are only known through SDP, for which neither RTP nor RTCP packets have been received, MUST NOT be counted for RTP group size estimation, and report blocks MUST NOT be sent for them in SR or RR RTCP messages. Endpoints MUST NOT assume that only the sources mentioned in SDP will be present in an RTP session; additional sources, with previously unmentioned SSRC IDs, can be added at any time, and endpoints MUST be prepared to receive packets from these sources. (How endpoints handle such packets is not specified here; they SHOULD be handled in the same manner as packets from additional sources would be handled had the endpoint not received any a=ssrc: attributes at all.) An endpoint that observes an SSRC collision between its explicitly-signaled source and another entity that has not explicitly signaled an SSRC MAY delay its RTP collision-resolution actions by 5*1.5*Td, where Td is the deterministic calculated reporting interval for receivers defined in Section 6.3.1 of the RTP specification, to see whether the conflict still exists. (This gives precedence to explicitly-signaled sources, and places the burden of collision resolution on non-signaled sources.) SSRC collisions between multiple explicitly-signaled sources, however, MUST be acted upon immediately. If, following RTP's collision-resolution procedures, a source identified by source-level attributes has been forced to change its SSRC identifier, the author of the SDP containing the source-level attributes for these sources SHOULD send out an updated SDP session description with the new SSRC, if the mechanism by which SDP is being distributed for the multimedia session has a mechanism to distribute updated SDP. This updated SDP MUST include a previous-ssrc source-level attribute, described in , listing the source's previous SSRC ID. (If only a single source with a given CNAME has collided, the other RTP session members can infer a correspondence between the source's old and new SSRC IDs, without requiring an updated session description. However, if more than one source collides at once, or if sources are leaving and re-joining, this inference is not possible. To avoid confusion, therefore, sending updated SDP messages is always RECOMMENDED.) Endpoints MUST NOT reuse the same SSRC ID for identified sources with same CNAME for at least the duration of the RTP session's participant timeout interval (see Section 6.3.5 of ). They SHOULD NOT reuse any SSRC ID ever mentioned in SDP (either by themselves or by other endpoints) for the entire lifetime of the RTP session. Endpoints MUST be prepared for the possibility that other parties in the session do not understand SDP source-level attributes, unless some higher-level mechanism normatively requires them. See for more discussion of this.

This section describes specific source attributes that can be applied to RTP sources.

cname: ]]>

The "cname" source attribute associates a media source with its Canonical End-Point Identifier (CNAME) source description (SDES) item. This MUST be the CNAME value that the media sender will place in its RTCP SDES packets; it therefore MUST follow the syntax conventions of CNAME defined in the RTP specification. If a session participant receives an RTCP SDES packet associating this SSRC with a different CNAME, it SHOULD assume there has been an SSRC collision, and that the description of the source that was carried in the SDP description is not applicable to the actual source being received. This source attribute is REQUIRED to be present if any source attributes are present for a source. The cname attribute MUST NOT occur more than once for the same ssrc-id within a given media stream. The "cname" source attribute is not dependent on charset. in gives a formal Augmented Backus-Naur Form (ABNF) grammar for the cname attribute.

previous-ssrc: ... ]]>

The "previous-ssrc" source attribute associates a media source with previous source identifiers used for the same media source. Following an SSRC change due to an SSRC collision involving a media source described in SDP, the updated session description describing the source's new SSRC (described in ) MUST include the previous-ssrc attribute associating the new SSRC with the old one. If further updated SDP descriptions are published describing the media source, the previous-ssrc attribute SHOULD be included if the session description was generated before the participant timeout of the old SSRC, and MAY be included after that point. This attribute, if present, MUST list at least one previous SSRC, and MAY list any number of additional SSRCs for the source, if the source has collided more than once. This attribute MUST be present only once for each source. The "previous-ssrc" source attribute is not dependent on charset. in gives a formal Augmented Backus-Naur Form (ABNF) grammar for the previous-ssrc attribute.

fmtp: ]]>

The "fmtp" source attribute allows format-specific parameters to be conveyed about a given source. The ]]> parameter MUST be one of the media formats (i.e., RTP payload types) specified for the media stream. The meaning of the ]]> is unique for each media type. This parameter MUST only be used for media types for which source-level format parameters have explicitly been specified; media-level format parameters MUST NOT be carried over blindly. The "fmtp" source attribute is not dependent on charset.

This document only defines source attributes which are necessary or useful for an endpoint to decode and render the sources in a media stream. It does include any attributes which would contribute to an endpoint's decision to accept or reject a stream, e.g. in an offer/answer exchange. Such attributes are for future consideration.

This section gives several examples of SDP descriptions of media sessions containing source attributes. For brevity, only the media sections of the descriptions are given.

m=audio 49168 RTP/AVP 0 a=ssrc:314159 cname:user@example.com

The example in shows an audio stream advertising a single source.

m=video 49170 RTP/AVP 96 a=rtpmap:96 H264/90000 a=ssrc:12345 cname:another-user@example.com a=ssrc:67890 cname:another-user@example.com

The example in shows a video stream where one participant (identified by a single CNAME) has several cameras. The sources could be further distinguished by RTCP Source Description (SDES) information.

m=video 49174 RTP/AVPF 96 98 a=rtpmap:96 H.264/90000 a=rtpmap:98 rtx/90000 a=fmtp:98 apt=96;rtx-time=3000 a=ssrc-group:FID 11111 22222 a=ssrc:11111 cname:user3@example.com a=ssrc:22222 cname:user3@example.com a=ssrc-group:FID 33333 44444 a=ssrc:33333 cname:user3@example.com a=ssrc:44444 cname:user3@example.com

The example in shows how the relationships among sources used for RTP Retransmission can be explicitly signaled. This prevents the complexity of associating original sources with retransmission sources when SSRC multiplexing is used for RTP retransmission, as is described in Section 5.3 of .

When used with the SDP Offer/Answer Model, SDP source-specific attributes describe only the sources with which each party is willing to send (whether it is sending RTP data or RTCP report blocks). No mechanism is provided by which an answer can accept or reject individual sources within a media stream; if the set of sources in a media stream is unacceptable, the answerer's only option is to reject the media stream or the entire multimedia session. The SSRC IDs for sources described by an SDP answer MUST be distinct from the SSRC IDs for sources of that media stream in the offer. Similarly, new SSRC IDs in an updated offer MUST be distinct from the ssrc IDs for that media stream established in the most recent offer/answer exchange for the session, and SHOULD be distinct from any SSRC ID ever used by either party within the multimedia session (whether or not it is still being used).

According to the defintion of SDP, interpreters of SDP session descriptions ignore unknown attributes. Thus, endpoints MUST be prepared that recipients of their RTP media session may not understand their explicit source descriptions, unless some external mechanism indicates that they were understood. In some cases (such as RTP Retransmission) this may constrain some choices about the bitstreams that are transmitted. Source descriptions are specified in this document such that RTP endpoints that are compliant with the RTP specification will be able to decode the media streams they describe whether or not they support explicit source descriptions. However, some deployed RTP implementations may not actually support multiple media sources in a media stream. Media senders MAY wish to restrict themselves to a single source at a time unless they have some means of concluding that the receivers of the media stream support source multiplexing.

This section gives a formal Augmented Backus-Naur Form (ABNF) grammar for each of the new media and source attributes defined in this document. Grammars for existing session or media attributes which have been extended to be source attributes are not included.

ssrc-attr = "ssrc:" ssrc-id SP attribute ; The base definition of "attribute" is in RFC 4566. ; (It is the content of "a=" lines.) ssrc-id = integer ; 0 - 2**32 - 1 attribute =/ ssrc-attr

ssrc-group-attr = "ssrc-group:" semantics *(SP ssrc-id) ; The definition of "semantics" is in RFC 3388. ; (It is the type of grouping being done.) attribute =/ ssrc-group-attr

cname-attr = "cname:" cname cname = byte-string ; Following the syntax conventions for CNAME as defined in RFC 3550. ; The definition of "byte-string" is in RFC 4566. attribute =/ cname-attr

previous-ssrc-attr = "previous-ssrc:" ssrc-id *(SP ssrc-id) attribute =/ previous-ssrc-attr

All the security implications of RTP and of SDP apply. Explicitly describing the multiplexed sources of an RTP media stream does not appear to add any further security issues.

This document defines two SDP media-level attributes: "ssrc" and "ssrc-group". These attributes should be registered by IANA under "Session Description Protocol (SDP) Parameters" under "att-field (media level only)". The "ssrc" attribute is used to identify characteristics of media sources within a media stream. Its format is defined in . The "ssrc-group" attribute is used to identify relationships among media sources within a media stream. Its format is defined in .

This specification creates a new IANA registry named "att-field (source level)" within the SDP parameters registry. Source attributes MUST be registered with IANA and documented, under the same rules as for SDP session-level and media-level attributes as specified in : New attribute registrations are accepted according to the "Specification Required" policy of , provided that the specification includes the following information: contact name, email address, and telephone number attribute name (as it will appear in SDP) long-form attribute name in English whether the attribute value is subject to the charset attribute a one-paragraph explanation of the purpose of the attribute a specification of appropriate attribute values for this attribute The above is the minimum that IANA will accept. Attributes that are expected to see widespread use and interoperability SHOULD be documented with a standards-track RFC that specifies the attribute more precisely. Submitters of registrations should ensure that the specification is in the spirit of SDP attributes, most notably that the attribute is platform independent in the sense that it makes no implicit assumptions about operating systems and does not name specific pieces of software in a manner that might inhibit interoperability. Source-level attributes which are substantially similar in semantics to existing session-level or media-level attributes SHOULD re-use the same attribute name as those session-level or media-level attributes. Source-level attributes SHOULD NOT re-use attribute names of session-level or media-level attributes that are unrelated or substantially different. The initial set of source attribute names, with definitions in of this document, is in .

Type SDP Name Reference ---- ------------------ --------- att-field (source level) cname [RFC&rfc.number;] previous-ssrc [RFC&rfc.number;] fmtp [RFC&rfc.number;]

(Note to the RFC-Editor: please replace "&rfc.number;" with the number of this document prior to publication as an RFC.)

This specification creates a new IANA registry named "Semantics for the "ssrc-group" SDP Attribute" within the SDP parameters registry. Source group semantics MUST be defined in standards-track RFCs, under the same rules as : 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. A brief description of the semantics. Token to be used within the group attribute. This token may be of any length, but SHOULD be no more than four characters long. Reference to an standards track RFC. Source grouping semantics values which are substantially similar to existing media grouping semantics values SHOULD re-use the same semantics name as that media gropuing semantics. Source grouping semantics SHOULD NOT re-use source grouping semantics names that are unrelated or substantially different. The initial set of source grouping semantics values, for the semantics specified in of this document, is in .

Semantics Token Reference ------------------- ----- --------- Flow Identification FID [RFC&rfc.number;] Forward Error Correction FEC [RFC&rfc.number;]

(Note to the RFC-Editor: please replace "&rfc.number;" with the number of this document prior to publication as an RFC.)

&abnf; &fec; &offeranswer; &rfc2119; &rfc5226; &rtp; &sdp; &sdpgroup; &anat; &flute_cs; &ice; &mikey; &mikeysdp; &rtprtx; &rtptopo; &srf; &ssm;

Note to the RFC-Editor: please remove this section prior to publication as an RFC.

Updated reference to RFC 2434 to .

Removed discussion of ssrc-multiplexing for layered codecs. Clarified that each "ssrc" attribute specifies only a single source-level attribute. Clarified that "ssrc-group" semantics are defined separately from "group" semantics. Clarified reference for the Td parameter. Updated references. Corrected typographical errors.

Added definitions of the new IANA registries and registrations needed. Clarified that none of the attributes defined in the document are dependent on the charset attribute. Clarified that ssrc attributes must be consistent with other SDP mechanisms (such as MIKEY) that also specify SSRCs. Removed open issues section.

Clarified that this document is expressly defining declarative source descriptions, not source offer/answer or other information. Removed the definitions of the "information", "bandwidth", "sendrecv", "sendonly", "recvonly", "inactive", "charset", "sdplang", "lang", "framerate", and "quality" source attributes. They are all unnecessary for source decodability, and to the extent they are otherwise useful they are probably better handled by RTCP Source Description (SDES) packets or feedback (AVPF) messages. Added text to the Backward Compatibility and Security Considerations sections.