Internet Engineering Task Force (IETF) C. Perkins Request for Comments: 5762 University of Glasgow Category: Standards Track April 2010 ISSN: 2070-1721
RTP and the Datagram Congestion Control Protocol (DCCP)
The Real-time Transport Protocol (RTP) is a widely used transport for real-time multimedia on IP networks. The Datagram Congestion Control Protocol (DCCP) is a transport protocol that provides desirable services for real-time applications. This memo specifies a mapping of RTP onto DCCP, along with associated signalling, such that real- time applications can make use of the services provided by DCCP.
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.
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.
Perkins Standards Track [Page 1]
RFC 5762 RTP over DCCP April 2010
This document may contain material from IETF Documents or IETF Contributions published or made publicly available before November 10, 2008. The person(s) controlling the copyright in some of this material may not have granted the IETF Trust the right to allow modifications of such material outside the IETF Standards Process. Without obtaining an adequate license from the person(s) controlling the copyright in such materials, this document may not be modified outside the IETF Standards Process, and derivative works of it may not be created outside the IETF Standards Process, except to format it for publication as an RFC or to translate it into languages other than English.
Table of Contents
1. Introduction ....................................................3 2. Rationale .......................................................3 3. Conventions Used in This Memo ...................................4 4. RTP over DCCP: Framing ..........................................4 4.1. RTP Data Packets ...........................................4 4.2. RTP Control Packets ........................................5 4.3. Multiplexing Data and Control ..............................7 4.4. RTP Sessions and DCCP Connections ..........................7 4.5. RTP Profiles ...............................................8 5. RTP over DCCP: Signalling using SDP .............................8 5.1. Protocol Identification ....................................8 5.2. Service Codes .............................................10 5.3. Connection Management .....................................11 5.4. Multiplexing Data and Control .............................11 5.5. Example ...................................................11 6. Security Considerations ........................................12 7. IANA Considerations ............................................13 8. Acknowledgements ...............................................14 9. References .....................................................14 9.1. Normative References ......................................14 9.2. Informative References ....................................15
The Real-time Transport Protocol (RTP)  is widely used in video streaming, telephony, and other real-time networked applications. RTP can run over a range of lower-layer transport protocols, and the performance of an application using RTP is heavily influenced by the choice of lower-layer transport. The Datagram Congestion Control Protocol (DCCP)  is a transport protocol that provides desirable properties for real-time applications running on unmanaged best- effort IP networks. This memo describes how RTP can be framed for transport using DCCP, and discusses some of the implications of such a framing. It also describes how the Session Description Protocol (SDP)  can be used to signal such sessions.
The remainder of this memo is structured as follows: it begins with a rationale for the work in Section 2, describing why a mapping of RTP onto DCCP is needed. Following a description of the conventions used in this memo in Section 3, the specification begins in Section 4 with the definition of how RTP packets are framed within DCCP. Associated signalling is described in Section 5. Security considerations are discussed in Section 6, and IANA considerations in Section 7.
With the widespread adoption of RTP have come concerns that many real-time applications do not implement congestion control, leading to the potential for congestion collapse of the network . The designers of RTP recognised this issue, stating in RFC 3551 that :
If best-effort service is being used, RTP receivers SHOULD monitor packet loss to ensure that the packet loss rate is within acceptable parameters. Packet loss is considered acceptable if a TCP flow across the same network path and experiencing the same network conditions would achieve an average throughput, measured on a reasonable timescale, that is not less than the RTP flow is achieving. This condition can be satisfied by implementing congestion control mechanisms to adapt the transmission rate (or the number of layers subscribed for a layered multicast session), or by arranging for a receiver to leave the session if the loss rate is unacceptably high.
While the goals are clear, the development of TCP friendly congestion control that can be used with RTP and real-time media applications is an open research question with many proposals for new algorithms, but little deployment experience.
Perkins Standards Track [Page 3]
RFC 5762 RTP over DCCP April 2010
Two approaches have been used to provide congestion control for RTP: 1) develop RTP extensions that incorporate congestion control; and 2) provide mechanisms for running RTP over congestion-controlled transport protocols. An example of the first approach can be found in , extending RTP to incorporate feedback information such that TCP Friendly Rate Control (TFRC)  can be implemented at the application level. This will allow congestion control to be added to existing applications without operating system or network support, and it offers the flexibility to experiment with new congestion control algorithms as they are developed. Unfortunately, it also passes the complexity of implementing congestion control onto application authors, a burden which many would prefer to avoid.
The second approach is to run RTP on a lower-layer transport protocol that provides congestion control. One possibility is to run RTP over TCP, as defined in , but the reliable nature of TCP and the dynamics of its congestion control algorithm make this inappropriate for most interactive real-time applications (the Stream Control Transmission Protocol (SCTP) is inappropriate for similar reasons). A better fit for such applications may be to run RTP over DCCP, since DCCP offers unreliable packet delivery and a choice of congestion control. This gives applications the ability to tailor the transport to their needs, taking advantage of better congestion control algorithms as they come available, while passing the complexity of implementation to the operating system. If DCCP should come to be widely available, it is believed these will be compelling advantages. Accordingly, this memo defines a mapping of RTP onto DCCP.
The following section defines how RTP and RTP Control Protocol (RTCP) packets can be framed for transport using DCCP. It also describes the differences between RTP sessions and DCCP connections, and the impact these have on the design of applications.
Each RTP data packet MUST be conveyed in a single DCCP datagram. Fields in the RTP header MUST be interpreted according to the RTP specification, and any applicable RTP Profile and Payload Format. Header processing is not affected by DCCP framing (in particular,
Perkins Standards Track [Page 4]
RFC 5762 RTP over DCCP April 2010
note that the semantics of the RTP sequence number and the DCCP sequence number are not compatible, and the value of one cannot be inferred from the other).
A DCCP connection is opened when an end system joins an RTP session, and it remains open for the duration of the session. To ensure NAT bindings are kept open, an end system SHOULD send a zero-length DCCP- Data packet once every 15 seconds during periods when it has no other data to send. This removes the need for RTP no-op packets , and similar application-level keepalives, when using RTP over DCCP. This application-level keepalive does not need to be sent if it is known that the DCCP CCID in use provides a transport-level keepalive, or if the application can determine that there are no NAT devices on the path.
RTP data packets MUST obey the dictates of DCCP congestion control. In some cases, the congestion control will require a sender to send at a rate below that which the payload format would otherwise use. To support this, an application could use either a rate-adaptive payload format, or a range of payload formats (allowing it to switch to a lower rate format if necessary). Details of the rate adaptation policy for particular payload formats are outside the scope of this memo (but see  and  for guidance).
RTP extensions that provide application-level congestion control (e.g., ) will conflict with DCCP congestion control, and MUST NOT be used.
DCCP allows an application to choose the checksum coverage, using a partial checksum to allow an application to receive packets with corrupt payloads. Some RTP Payload Formats (e.g., ) can make use of this feature in conjunction with payload-specific mechanisms to improve performance when operating in environments with frequent non- congestive packet corruption. If such a payload format is used, an RTP end system MAY enable partial checksums at the DCCP layer, in which case the checksum MUST cover at least the DCCP and RTP headers to ensure packets are correctly delivered. Partial checksums MUST NOT be used unless supported by mechanisms in the RTP payload format.
The RTP Control Protocol (RTCP) is used in the standard manner with DCCP. RTCP packets are grouped into compound packets, as described in Section 6.1 of , and each compound RTCP packet is transported in a single DCCP datagram.
Perkins Standards Track [Page 5]
RFC 5762 RTP over DCCP April 2010
The usual RTCP timing rules apply, with the additional constraint that RTCP packets MUST obey the DCCP congestion control algorithm negotiated for the connection. This can prevent a participant from sending an RTCP packet at the expiration of the RTCP transmission timer if there is insufficient network capacity available. In such cases the RTCP packet is delayed and sent at the earliest possible instant when capacity becomes available. The actual time the RTCP packet was sent is then used as the basis for calculating the next RTCP transmission time.
RTCP packets comprise only a small fraction of the total traffic in an RTP session. Accordingly, it is expected that delays in their transmission due to congestion control will not be common, provided the configured nominal "session bandwidth" (see Section 6.2 of ) is in line with the bandwidth achievable on the DCCP connection. If, however, the capacity of the DCCP connection is significantly below the nominal session bandwidth, RTCP packets may be delayed enough for participants to time out due to apparent inactivity. In such cases, the session parameters SHOULD be re-negotiated to more closely match the available capacity, for example by performing a re-invite with an updated "b=" line when using the Session Initiation Protocol  for signalling.
Note: Since the nominal session bandwidth is chosen based on media codec capabilities, a session where the nominal bandwidth is much larger than the available bandwidth will likely become unusable due to constraints on the media channel, and so require negotiation of a lower bandwidth codec, before it becomes unusable due to constraints on the RTCP channel.
As noted in Section 17.1 of , there is the potential for overlap between information conveyed in RTCP packets and that conveyed in DCCP acknowledgement options. In general this is not an issue since RTCP packets contain media-specific data that is not present in DCCP acknowledgement options, and DCCP options contain network-level data that is not present in RTCP. Indeed, there is no overlap between the five RTCP packet types defined in the RTP specification  and the standard DCCP options . There are, however, cases where overlap does occur: most clearly between the Loss RLE Report Blocks defined as part of the RTCP Extended Reports  and the DCCP Ack Vector option. If there is overlap between RTCP report packets and DCCP acknowledgements, an application SHOULD use either RTCP feedback or DCCP acknowledgements, but not both (use of both types of feedback will waste available network capacity, but is not otherwise harmful).
The obvious mapping of RTP onto DCCP creates two DCCP connections for each RTP flow: one for RTP data packets and one for RTP control packets. A frequent criticism of RTP relates to the number of ports it uses, since large telephony gateways can support more than 32768 RTP flows between pairs of gateways, and so run out of UDP ports. In addition, use of multiple ports complicates NAT traversal. For these reasons, it is RECOMMENDED that the RTP and RTCP traffic for a single RTP session is multiplexed onto a single DCCP connection following the guidelines in , where possible (it may not be possible in all circumstances, for example when translating from an RTP stream over a non-DCCP transport that uses conflicting RTP payload types and RTCP packet types).
An end system SHOULD NOT assume that it will observe only a single RTP synchronisation source (SSRC) because it is using DCCP framing. An RTP session can span any number of transport connections, and can include RTP mixers or translators bringing other participants into the session. The use of a unicast DCCP connection does not imply that the RTP session will have only two participants, and RTP end systems SHOULD assume that multiple synchronisation sources may be observed when using RTP over DCCP, unless otherwise signalled.
An RTP translator bridging multiple DCCP connections to form a single RTP session needs to be aware of the congestion state of each DCCP connection, and must adapt the media to the available capacity of each. The Codec Control Messages defined in  may be used to signal congestion state to the media senders, allowing them to adapt their transmission. Alternatively, media transcoding may be used to perform adaptation: this is computationally expensive, induces delay, and generally gives poor-quality results. Depending on the payload, it might also be possible to use some form of scalable coding.
A single RTP session may also span a DCCP connection and some other type of transport connection. An example might be an RTP over DCCP connection from an RTP end system to an RTP translator, with an RTP over UDP/IP multicast group on the other side of the translator. A second example might be an RTP over DCCP connection that links Public Switched Telephone Network (PSTN) gateways. The issues for such an RTP translator are similar to those when linking two DCCP connections, except that the congestion control algorithms on either side of the translator may not be compatible. Implementation of effective translators for such an environment is non-trivial.
In general, there is no conflict between new RTP profiles and DCCP framing, and most RTP profiles can be negotiated for use over DCCP with the following exceptions:
o An RTP profile that is intolerant of packet corruption may conflict with the DCCP partial checksum feature. An example of this is the integrity protection provided by the RTP/SAVP profile, which cannot be used in conjunction with DCCP partial checksums.
o An RTP profile that mandates a particular non-DCCP lower-layer transport will conflict with DCCP.
RTP profiles that fall under these exceptions SHOULD NOT be used with DCCP unless the conflicting features can be disabled.
Of the profiles currently defined, the RTP Profile for Audio and Video Conferences with Minimal Control , the Secure Real-time Transport Protocol , the Extended RTP Profile for RTCP-based Feedback , and the Extended Secure RTP Profile for RTCP-based Feedback  MAY be used with DCCP (noting the potential conflict between DCCP partial checksums and the integrity protection provided by the secure RTP variants -- see Section 6).
The Session Description Protocol (SDP)  and the offer/answer model  are widely used to negotiate RTP sessions (for example, using the Session Initiation Protocol ). This section describes how SDP is used to signal RTP sessions running over DCCP.
SDP uses a media ("m=") line to convey details of the media format and transport protocol used. The ABNF syntax of a media line is as follows (from ):
media-field = %x6d "=" media SP port ["/" integer] SP proto 1*(SP fmt) CRLF
The proto field denotes the transport protocol used for the media, while the port indicates the transport port to which the media is sent. Following  and , this memo defines these five values of the proto field to indicate media transported using DCCP:
The "DCCP" protocol identifier is similar to the "UDP" and "TCP" protocol identifiers and denotes the DCCP transport protocol , but not its upper-layer protocol. An SDP "m=" line that specifies the "DCCP" protocol MUST further qualify the application-layer protocol using a "fmt" identifier (the "fmt" namespace is managed in the same manner as for the "UDP" protocol identifier). A single DCCP port is used, as denoted by the port field in the media line. The "DCCP" protocol identifier MUST NOT be used to signal RTP sessions running over DCCP; those sessions MUST use a protocol identifier of the form "DCCP/RTP/..." as described below.
The "DCCP/RTP/AVP" protocol identifier refers to RTP using the RTP Profile for Audio and Video Conferences with Minimal Control  running over DCCP.
The "DCCP/RTP/SAVP" protocol identifier refers to RTP using the Secure Real-time Transport Protocol  running over DCCP.
The "DCCP/RTP/AVPF" protocol identifier refers to RTP using the Extended RTP Profile for RTCP-based Feedback  running over DCCP.
The "DCCP/RTP/SAVPF" protocol identifier refers to RTP using the Extended Secure RTP Profile for RTCP-based Feedback  running over DCCP.
RTP payload formats used with the "DCCP/RTP/AVP", "DCCP/RTP/SAVP", "DCCP/RTP/AVPF", and "DCCP/RTP/SAVPF" protocol identifiers MUST use the payload type number as their "fmt" value. If the payload type number is dynamically assigned, an additional "rtpmap" attribute MUST be included to specify the format name and parameters as defined by the media type registration for the payload format.
DCCP port 5004 is registered for use by the RTP profiles listed above, and SHOULD be the default port chosen by applications using those profiles. If multiple RTP sessions are active from a host, even-numbered ports in the dynamic range SHOULD be used for the other sessions. If RTCP is to be sent on a separate DCCP connection to RTP, the RTCP connection SHOULD use the next higher destination port number, unless an alternative DCCP port is signalled using the "a=rtcp:" attribute . For improved interoperability, "a=rtcp:" SHOULD be used whenever an alternate DCCP port is used.
In addition to the port number, specified on the SDP "m=" line, a DCCP connection has an associated service code. A single new SDP attribute ("dccp-service-code") is defined to signal the DCCP service code according to the following ABNF :
where DIGIT and HEXDIG are as defined in . The service code is interpreted as defined in Section 8.1.2 of  and may be specified using either the hexadecimal, decimal, or ASCII formats. A parser MUST interpret service codes according to their numeric value, independent of the format used to represent them in SDP.
The following DCCP service codes are registered for use with RTP:
o SC:RTPA (equivalently SC=1381257281 or SC=x52545041): an RTP session conveying audio data (and OPTIONAL multiplexed RTCP)
o SC:RTPV (equivalently SC=1381257302 or SC=x52545056): an RTP session conveying video data (and OPTIONAL multiplexed RTCP)
o SC:RTPT (equivalently SC=1381257300 or SC=x52545054): an RTP session conveying text media (and OPTIONAL multiplexed RTCP)
o SC:RTPO (equivalently SC=1381257295 or SC=x5254504f): an RTP session conveying any other type of media (and OPTIONAL multiplexed RTCP)
o SC:RTCP (equivalently SC=1381253968 or SC=x52544350): an RTCP connection, separate from the corresponding RTP
To ease the job of middleboxes, applications SHOULD use these service codes to identify RTP sessions running within DCCP. The service code SHOULD match the top-level media type signalled for the session
Perkins Standards Track [Page 10]
RFC 5762 RTP over DCCP April 2010
(i.e., the SDP "m=" line), with the exception connections using media types other than audio, video, or text, which use SC:RTPO, and connections that transport only RTCP packets, which use SC:RTCP.
The "a=dccp-service-code:" attribute is a media-level attribute that is not subject to the charset attribute.
The "a=setup:" attribute indicates which of the endpoints should initiate the DCCP connection establishment (i.e., send the initial DCCP-Request packet). The "a=setup:" attribute MUST be used in a manner comparable with , except that DCCP connections are being initiated rather than TCP connections.
After the initial offer/answer exchange, the endpoints may decide to re-negotiate various parameters. The "a=connection:" attribute MUST be used in a manner compatible with  to decide whether a new DCCP connection needs to be established as a result of subsequent offer/ answer exchanges, or if the existing connection should still be used.
A single DCCP connection can be used to transport multiplexed RTP and RTCP packets. Such multiplexing MUST be signalled using an "a=rtcp- mux" attribute according to . If multiplexed RTP and RTCP are not to be used, then the "a=rtcp-mux" attribute MUST NOT be present in the SDP offer, and a separate DCCP connection MUST be opened to transport the RTCP data on a different DCCP port.
An offerer at 192.0.2.47 signals its availability for an H.261 video session, using RTP/AVP over DCCP with service code "RTPV" (using the hexadecimal encoding of the service code in the SDP). RTP and RTCP packets are multiplexed onto a single DCCP connection:
The end point at 192.0.2.128 then initiates a DCCP connection to port 5004 at 192.0.2.47. DCCP port 5004 is used for both the RTP and RTCP data, and port 5005 is unused. The textual encoding of the service code is used in the answer, and represents the same service code as in the offer.
The security considerations in the RTP specification  and any applicable RTP profile (e.g., , , , or ) or payload format apply when transporting RTP over DCCP.
The security considerations in the DCCP specification  apply.
The SDP signalling described in Section 5 is subject to the security considerations of , , , , and .
The provision of effective congestion control for RTP through use of DCCP is expected to help reduce the potential for denial of service present when RTP flows ignore the advice in  to monitor packet loss and reduce their sending rate in the face of persistent congestion.
There is a potential conflict between the Secure RTP profiles (, ) and the DCCP partial checksum option, since these profiles introduce, and recommend the use of, message authentication for RTP and RTCP packets. Message authentication codes of the type used by these profiles cannot be used with partial checksums, since any bit error in the DCCP packet payload will cause the authentication check to fail. Accordingly, DCCP partial checksums SHOULD NOT be used in conjunction with Secure Real-time Transport Protocol (SRTP) authentication. The confidentiality features of the basic RTP specification cannot be used with DCCP partial checksums, since bit
Perkins Standards Track [Page 12]
RFC 5762 RTP over DCCP April 2010
errors propagate. Also, despite the fact that bit errors do not propagate when using AES in counter mode, the Secure RTP profiles SHOULD NOT be used with DCCP partial checksums, since the profiles require authentication for security, and authentication is incompatible with partial checksums.
Note: ports 5004/tcp, 5004/udp, 5005/tcp, and 5005/udp have existing registrations, but incorrect descriptions and references. The IANA has updated the existing registrations as follows:
avt-profile-1 5004/tcp RTP media data [RFC3551, RFC4571] avt-profile-1 5004/udp RTP media data [RFC3551] avt-profile-2 5005/tcp RTP control protocol [RFC3551, RFC4571] avt-profile-2 5005/udp RTP control protocol [RFC3551]
This work was supported in part by the UK Engineering and Physical Sciences Research Council. Thanks are due to Philippe Gentric, Magnus Westerlund, Sally Floyd, Dan Wing, Gorry Fairhurst, Stephane Bortzmeyer, Arjuna Sathiaseelan, Tom Phelan, Lars Eggert, Eddie Kohler, Miguel Garcia, and the other members of the DCCP working group for their comments.
 Floyd, S. and J. Kempf, "IAB Concerns Regarding Congestion Control for Voice Traffic in the Internet", RFC 3714, March 2004.
 Gharai, L., "RTP with TCP Friendly Rate Control", Work in Progress, July 2007.
 Floyd, S., Handley, M., Padhye, J., and J. Widmer, "TCP Friendly Rate Control (TFRC): Protocol Specification", RFC 5348, September 2008.
 Andreasen, F., Oran, D., and D. Wing, "A No-Op Payload Format for RTP", Work in Progress, May 2005.
 Phelan, T., "Strategies for Streaming Media Applications Using TCP-Friendly Rate Control", Work in Progress, July 2007.
 Phelan, T., "Datagram Congestion Control Protocol (DCCP) User Guide", Work in Progress, April 2005.
Perkins Standards Track [Page 15]
RFC 5762 RTP over DCCP April 2010
 Sjoberg, J., Westerlund, M., Lakaniemi, A., and Q. Xie, "RTP Payload Format and File Storage Format for the Adaptive Multi- Rate (AMR) and Adaptive Multi-Rate Wideband (AMR-WB) Audio Codecs", RFC 4867, April 2007.
 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.
 Friedman, T., Caceres, R., and A. Clark, "RTP Control Protocol Extended Reports (RTCP XR)", RFC 3611, November 2003.
 Wenger, S., Chandra, U., Westerlund, M., and B. Burman, "Codec Control Messages in the RTP Audio-Visual Profile with Feedback (AVPF)", Work in Progress, October 2007.
Colin Perkins University of Glasgow Department of Computing Science Glasgow G12 8QQ UK