Web Real-Time Communication (WebRTC): Media
Transport and Use of RTPUniversity of GlasgowSchool of Computing ScienceGlasgowG12 8QQUnited Kingdomcsp@csperkins.orghttp://csperkins.org/EricssonFarogatan 6SE-164 80 KistaSweden+46 10 714 82 87magnus.westerlund@ericsson.comAalto UniversitySchool of Electrical EngineeringEspoo02150Finlandjorg.ott@aalto.fiRTCWEB Working GroupThe Web Real-Time Communication (WebRTC) framework provides support
for direct interactive rich communication using audio, video, text,
collaboration, games, etc. between two peers' web-browsers. This memo
describes the media transport aspects of the WebRTC framework. It
specifies how the Real-time Transport Protocol (RTP) is used in the
WebRTC context, and gives requirements for which RTP features, profiles,
and extensions need to be supported.The Real-time Transport Protocol (RTP)
provides a framework for delivery of audio and video teleconferencing
data and other real-time media applications. Previous work has defined
the RTP protocol, along with numerous profiles, payload formats, and
other extensions. When combined with appropriate signalling, these form
the basis for many teleconferencing systems.The Web Real-Time communication (WebRTC) framework provides the
protocol building blocks to support direct, interactive, real-time
communication using audio, video, collaboration, games, etc., between
two peers' web-browsers. This memo describes how the RTP framework is to
be used in the WebRTC context. It proposes a baseline set of RTP
features that are to be implemented by all WebRTC-aware end-points,
along with suggested extensions for enhanced functionality.This memo specifies a protocol intended for use within the WebRTC
framework, but is not restricted to that context. An overview of the
WebRTC framework is given in .The structure of this memo is as follows. outlines our rationale in preparing this memo
and choosing these RTP features.
defines terminology. Requirements for core RTP protocols are described
in and suggested RTP extensions are
described in . outlines mechanisms that can increase
robustness to network problems, while
describes congestion control and rate adaptation mechanisms. The
discussion of mandated RTP mechanisms concludes in with a review of performance monitoring and network
management tools that can be used in the WebRTC context. gives some guidelines for future incorporation of
other RTP and RTP Control Protocol (RTCP) extensions into this
framework. describes requirements placed
on the signalling channel. discusses the
relationship between features of the RTP framework and the WebRTC
application programming interface (API), and discusses RTP implementation considerations. The
memo concludes with security
considerations and IANA
considerations.The RTP framework comprises the RTP data transfer protocol, the RTP
control protocol, and numerous RTP payload formats, profiles, and
extensions. This range of add-ons has allowed RTP to meet various needs
that were not envisaged by the original protocol designers, and to
support many new media encodings, but raises the question of what
extensions are to be supported by new implementations. The development
of the WebRTC framework provides an opportunity to review the available
RTP features and extensions, and to define a common baseline feature set
for all WebRTC implementations of RTP. This builds on the past 20 years
development of RTP to mandate the use of extensions that have shown
widespread utility, while still remaining compatible with the wide
installed base of RTP implementations where possible.RTP and RTCP extensions that are not discussed in this document can
be implemented by WebRTC end-points if they are beneficial for new use
cases. However, they are not necessary to address the WebRTC use cases
and requirements identified in .While the baseline set of RTP features and extensions defined in this
memo is targeted at the requirements of the WebRTC framework, it is
expected to be broadly useful for other conferencing-related uses of
RTP. In particular, it is likely that this set of RTP features and
extensions will be appropriate for other desktop or mobile video
conferencing systems, or for room-based high-quality telepresence
applications.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 .
The RFC 2119 interpretation of these key words applies only when written
in ALL CAPS. Lower- or mixed-case uses of these key words are not to be
interpreted as carrying special significance in this memo.We define the following additional terms:The MediaStream concept defined by
the W3C in the WebRTC API.A uni-directional flow of
transport packets that are identified by having a particular 5-tuple
of source IP address, source port, destination IP address,
destination port, and transport protocol used.A bi-directional
transport-layer flow is a transport-layer flow that is symmetric.
That is, the transport-layer flow in the reverse direction has a
5-tuple where the source and destination address and ports are
swapped compared to the forward path transport-layer flow, and the
transport protocol is the same.This document uses the terminology from . Other terms are used
according to their definitions from the RTP
Specification. We especially note the following frequently used
terms: RTP Packet Stream, RTP Session, and End-point.The following sections describe the core features of RTP and RTCP
that need to be implemented, along with the mandated RTP profiles. Also
described are the core extensions providing essential features that all
WebRTC implementations need to implement to function effectively on
today's networks.The Real-time Transport Protocol (RTP)
is REQUIRED to be implemented as the media transport protocol
for WebRTC. RTP itself comprises two parts: the RTP data transfer
protocol, and the RTP control protocol (RTCP). RTCP is a fundamental
and integral part of RTP, and MUST be implemented in all WebRTC
applications.The following RTP and RTCP features are sometimes omitted in
limited functionality implementations of RTP, but are REQUIRED in all
WebRTC implementations: Support for use of multiple simultaneous SSRC values in a
single RTP session, including support for RTP end-points that send
many SSRC values simultaneously, following and . Support for the RTCP
optimisations for multi-SSRC sessions defined in is
RECOMMENDED.Random choice of SSRC on joining a session; collision detection
and resolution for SSRC values (see also ).Support for reception of RTP data packets containing CSRC
lists, as generated by RTP mixers, and RTCP packets relating to
CSRCs.Sending correct synchronisation information in the RTCP Sender
Reports, to allow receivers to implement lip-synchronisation;
support for the rapid RTP synchronisation extensions (see ) is RECOMMENDED.Support for multiple synchronisation contexts. Participants
that send multiple simultaneous RTP packet streams SHOULD do so as
part of a single synchronisation context, using a single RTCP
CNAME for all streams and allowing receivers to play the streams
out in a synchronised manner. For compatibility with potential
future versions of this specification, or for interoperability
with non-WebRTC devices through a gateway, receivers MUST support
multiple synchronisation contexts, indicated by the use of
multiple RTCP CNAMEs in an RTP session. This specification
requires the usage of a single CNAME when sending RTP Packet
Streams in some circumstances, see .Support for sending and receiving RTCP SR, RR, SDES, and BYE
packet types, with OPTIONAL support for other RTCP packet types
unless mandated by other parts of this specification;
implementations MUST ignore unknown RTCP packet types. Note that
additional RTCP Packet types are used by the RTP/SAVPF Profile and the other RTCP extensions.Support for multiple end-points in a single RTP session, and
for scaling the RTCP transmission interval according to the number
of participants in the session; support for randomised RTCP
transmission intervals to avoid synchronisation of RTCP reports;
support for RTCP timer reconsideration.Support for configuring the RTCP bandwidth as a fraction of the
media bandwidth, and for configuring the fraction of the RTCP
bandwidth allocated to senders, e.g., using the SDP "b=" line
. Support for the
reduced minimum RTCP reporting interval described in Section 6.2
of is RECOMMENDED.It is known that a significant number of legacy RTP
implementations, especially those targeted at VoIP-only systems, do
not support all of the above features, and in some cases do not
support RTCP at all. Implementers are advised to consider the
requirements for graceful degradation when interoperating with legacy
implementations.Other implementation considerations are discussed in .The complete specification of RTP for a particular application
domain requires the choice of an RTP Profile. For WebRTC use, the
Extended Secure RTP Profile for RTCP-Based
Feedback (RTP/SAVPF), as extended by ,
MUST be implemented. The RTP/SAVPF profile is the combination of basic
RTP/AVP profile, the RTP profile for RTCP-based feedback
(RTP/AVPF), and the secure RTP profile
(RTP/SAVP).The RTCP-based feedback extensions are
needed for the improved RTCP timer model. This allows more flexible
transmission of RTCP packets in response to events, rather than
strictly according to bandwidth, and is vital for being able to report
congestion signals as well as media events. These extensions also
allow saving RTCP bandwidth, and an end-point will commonly only use
the full RTCP bandwidth allocation if there are many events that
require feedback. The timer rules are also needed to make use of the
RTP conferencing extensions discussed in .Note: The enhanced RTCP timer model defined in the RTP/AVPF
profile is backwards compatible with legacy systems that implement
only the RTP/AVP or RTP/SAVP profile, given some constraints on
parameter configuration such as the RTCP bandwidth value and
"trr-int" (the most important factor for interworking with
RTP/(S)AVP end-points via a gateway is to set the trr-int
parameter to a value representing 4 seconds).The secure RTP (SRTP) profile extensions
are needed to provide media encryption, integrity protection, replay
protection and a limited form of source authentication. WebRTC
implementations MUST NOT send packets using the basic RTP/AVP profile
or the RTP/AVPF profile; they MUST employ the full RTP/SAVPF profile
to protect all RTP and RTCP packets that are generated (i.e.,
implementations MUST use SRTP and SRTCP). The RTP/SAVPF profile MUST
be configured using the cipher suites, DTLS-SRTP protection profiles,
keying mechanisms, and other parameters described in .The set of mandatory to implement codecs and RTP payload formats
for WebRTC is not specified in this memo, instead they are defined in
separate specifications, such as . Implementations can support any
codec for which an RTP payload format and associated signalling is
defined. Implementation cannot assume that the other participants in
an RTP session understand any RTP payload format, no matter how
common; the mapping between RTP payload type numbers and specific
configurations of particular RTP payload formats MUST be agreed before
those payload types/formats can be used. In an SDP context, this can
be done using the "a=rtpmap:" and "a=fmtp:" attributes associated with
an "m=" line, along with any other SDP attributes needed to configure
the RTP payload format.End-points can signal support for multiple RTP payload formats, or
multiple configurations of a single RTP payload format, as long as
each unique RTP payload format configuration uses a different RTP
payload type number. As outlined in , the RTP
payload type number is sometimes used to associate an RTP packet
stream with a signalling context. This association is possible
provided unique RTP payload type numbers are used in each context. For
example, an RTP packet stream can be associated with an SDP "m=" line
by comparing the RTP payload type numbers used by the RTP packet
stream with payload types signalled in the "a=rtpmap:" lines in the
media sections of the SDP. If RTP packet streams are being associated
with signalling contexts based on the RTP payload type, then the
assignment of RTP payload type numbers MUST be unique across
signalling contexts; if the same RTP payload format configuration is
used in multiple contexts, then a different RTP payload type number
has to be assigned in each context to ensure uniqueness. If the RTP
payload type number is not being used to associate RTP packet streams
with a signalling context, then the same RTP payload type number can
be used to indicate the exact same RTP payload format configuration in
multiple contexts. A single RTP payload type number MUST NOT be
assigned to different RTP payload formats, or different configurations
of the same RTP payload format, within a single RTP session (note that
the different "m=" lines in an SDP bundle
group form a single RTP session).An end-point that has signalled support for multiple RTP payload
formats SHOULD be able to accept data in any of those payload formats
at any time, unless it has previously signalled limitations on its
decoding capability. This requirement is constrained if several types
of media (e.g., audio and video) are sent in the same RTP session. In
such a case, a source (SSRC) is restricted to switching only between
the RTP payload formats signalled for the type of media that is being
sent by that source; see . To support
rapid rate adaptation by changing codec, RTP does not require advance
signalling for changes between RTP payload formats used by a single
SSRC that were signalled during session set-up.An RTP sender that changes between two RTP payload types that use
different RTP clock rates MUST follow the recommendations in Section
4.1 of . RTP receivers MUST follow the
recommendations in Section 4.3 of in order to
support sources that switch between clock rates in an RTP session
(these recommendations for receivers are backwards compatible with the
case where senders use only a single clock rate).An association amongst a set of end-points communicating using RTP
is known as an RTP session . An end-point can
be involved in several RTP sessions at the same time. In a multimedia
session, each type of media has typically been carried in a separate
RTP session (e.g., using one RTP session for the audio, and a separate
RTP session using a different transport-layer flow for the video).
WebRTC implementations of RTP are REQUIRED to implement support for
multimedia sessions in this way, separating each session using
different transport-layer flows for compatibility with legacy
systems.In modern day networks, however, with the widespread use of network
address/port translators (NAT/NAPT) and firewalls, it is desirable to
reduce the number of transport-layer flows used by RTP applications.
This can be done by sending all the RTP packet streams in a single RTP
session, which will comprise a single transport-layer flow (this will
prevent the use of some quality-of-service mechanisms, as discussed in
). Implementations are therefore
also REQUIRED to support transport of all RTP packet streams,
independent of media type, in a single RTP session using a single
transport layer flow, according to . If multiple types
of media are to be used in a single RTP session, all participants in
that RTP session MUST agree to this usage. In an SDP context, can be used to
signal such a bundle of RTP packet streams forming a single RTP
session.Further discussion about the suitability of different RTP session
structures and multiplexing methods to different scenarios are
suitable can be found in .Historically, RTP and RTCP have been run on separate transport
layer flows (e.g., two UDP ports for each RTP session, one port for
RTP and one port for RTCP). With the increased use of Network
Address/Port Translation (NAT/NAPT) this has become problematic, since
maintaining multiple NAT bindings can be costly. It also complicates
firewall administration, since multiple ports need to be opened to
allow RTP traffic. To reduce these costs and session set-up times,
support for multiplexing RTP data packets and RTCP control packets on
a single transport-layer flow for each RTP session is REQUIRED,
provided it is negotiated in the signalling channel before use as
specified in . For backwards compatibility,
implementations are also REQUIRED to support RTP and RTCP sent on
separate transport-layer flows.Note that the use of RTP and RTCP multiplexed onto a single
transport-layer flow ensures that there is occasional traffic sent on
that port, even if there is no active media traffic. This can be
useful to keep NAT bindings alive, and is the recommend method for
application level keep-alives of RTP
sessions.RTCP packets are usually sent as compound RTCP packets, and requires that those compound packets start with an
Sender Report (SR) or Receiver Report (RR) packet. When using frequent
RTCP feedback messages under the RTP/AVPF Profile these statistics are not needed in every packet,
and unnecessarily increase the mean RTCP packet size. This can limit
the frequency at which RTCP packets can be sent within the RTCP
bandwidth share.To avoid this problem, specifies how to
reduce the mean RTCP message size and allow for more frequent
feedback. Frequent feedback, in turn, is essential to make real-time
applications quickly aware of changing network conditions, and to
allow them to adapt their transmission and encoding behaviour. Support
for non-compound RTCP feedback packets is
REQUIRED, but MUST be negotiated using the signalling channel before
use. For backwards compatibility, implementations are also REQUIRED to
support the use of compound RTCP feedback packets if the remote
end-point does not agree to the use of non-compound RTCP in the
signalling exchange.To ease traversal of NAT and firewall devices, implementations are
REQUIRED to implement and use Symmetric
RTP. The reason for using symmetric RTP is primarily to avoid
issues with NATs and Firewalls by ensuring that the send and receive
RTP packet streams, as well as RTCP, are actually bi-directional
transport-layer flows. This will keep alive the NAT and firewall
pinholes, and help indicate consent that the receive direction is a
transport-layer flow the intended recipient actually wants. In
addition, it saves resources, specifically ports at the end-points,
but also in the network as NAT mappings or firewall state is not
unnecessary bloated. The amount of per flow QoS state kept in the
network is also reduced.Implementations are REQUIRED to support signalled RTP
synchronisation source (SSRC) identifiers, using the "a=ssrc:" SDP
attribute defined in Section 4.1 and Section 5 of . Implementations MUST also support the
"previous-ssrc" source attribute defined in Section 6.2 of . Other per-SSRC attributes defined in MAY be supported.Use of the "a=ssrc:" attribute to signal SSRC identifiers in an RTP
session is OPTIONAL. Implementations MUST be prepared to accept RTP
and RTCP packets using SSRCs that have not been explicitly signalled
ahead of time. Implementations MUST support random SSRC assignment,
and MUST support SSRC collision detection and resolution, according to
. When using signalled SSRC values, collision
detection MUST be performed as described in Section 5 of .It is often desirable to associate an RTP packet stream with a
non-RTP context. For users of the WebRTC API a mapping between SSRCs
and MediaStreamTracks are provided per . For gateways or other usages it is possible
to associate an RTP packet stream with an "m=" line in a session
description formatted using SDP. If SSRCs are signalled this is
straightforward (in SDP the "a=ssrc:" line will be at the media level,
allowing a direct association with an "m=" line). If SSRCs are not
signalled, the RTP payload type numbers used in an RTP packet stream
are often sufficient to associate that packet stream with a signalling
context (e.g., if RTP payload type numbers are assigned as described
in of this memo, the RTP payload types
used by an RTP packet stream can be compared with values in SDP
"a=rtpmap:" lines, which are at the media level in SDP, and so map to
an "m=" line).The RTCP Canonical Name (CNAME) provides a persistent
transport-level identifier for an RTP end-point. While the
Synchronisation Source (SSRC) identifier for an RTP end-point can
change if a collision is detected, or when the RTP application is
restarted, its RTCP CNAME is meant to stay unchanged for the duration
of a RTCPeerConnection,
so that RTP end-points can be uniquely identified and associated with
their RTP packet streams within a set of related RTP sessions.Each RTP end-point MUST have at least one RTCP CNAME, and that RTCP
CNAME MUST be unique within the RTCPeerConnection. RTCP CNAMEs
identify a particular synchronisation context, i.e., all SSRCs
associated with a single RTCP CNAME share a common reference clock. If
an end-point has SSRCs that are associated with several unsynchronised
reference clocks, and hence different synchronisation contexts, it
will need to use multiple RTCP CNAMEs, one for each synchronisation
context.Taking the discussion in into
account, a WebRTC end-point MUST NOT use more than one RTCP CNAME in
the RTP sessions belonging to single RTCPeerConnection (that is, an
RTCPeerConnection forms a synchronisation context). RTP middleboxes
MAY generate RTP packet streams associated with more than one RTCP
CNAME, to allow them to avoid having to resynchronize media from
multiple different end-points part of a multi-party RTP session.The RTP specification includes
guidelines for choosing a unique RTP CNAME, but these are not
sufficient in the presence of NAT devices. In addition, long-term
persistent identifiers can be problematic from a privacy viewpoint. Accordingly, a WebRTC
endpoint MUST generate a new, unique, short-term persistent RTCP CNAME
for each RTCPeerConnection, following , with a
single exception; if explicitly requested at creation an
RTCPeerConnection MAY use the same CNAME as as an existing
RTCPeerConnection within their common same-origin context.An WebRTC end-point MUST support reception of any CNAME that
matches the syntax limitations specified by the RTP specification and cannot assume that any
CNAME will be chosen according to the form suggested above.The guidelines regarding handling of leap seconds to limit their
impact on RTP media playout and synchronization given in SHOULD be followed.There are a number of RTP extensions that are either needed to obtain
full functionality, or extremely useful to improve on the baseline
performance, in the WebRTC application context. One set of these
extensions is related to conferencing, while others are more generic in
nature. The following subsections describe the various RTP extensions
mandated or suggested for use within the WebRTC context.RTP is a protocol that inherently supports group communication.
Groups can be implemented by having each endpoint send its RTP packet
streams to an RTP middlebox that redistributes the traffic, by using a
mesh of unicast RTP packet streams between endpoints, or by using an
IP multicast group to distribute the RTP packet streams. These
topologies can be implemented in a number of ways as discussed in
.While the use of IP multicast groups is popular in IPTV systems,
the topologies based on RTP middleboxes are dominant in interactive
video conferencing environments. Topologies based on a mesh of unicast
transport-layer flows to create a common RTP session have not seen
widespread deployment to date. Accordingly, WebRTC implementations are
not expected to support topologies based on IP multicast groups or to
support mesh-based topologies, such as a point-to-multipoint mesh
configured as a single RTP session (Topo-Mesh in the terminology of
). However, a
point-to-multipoint mesh constructed using several RTP sessions,
implemented in the WebRTC context using independent
RTCPeerConnections, can be expected to be utilised by WebRTC
applications and needs to be supported.WebRTC implementations of RTP endpoints implemented according to
this memo are expected to support all the topologies described in
where the RTP
endpoints send and receive unicast RTP packet streams to and from some
peer device, provided that peer can participate in performing
congestion control on the RTP packet streams. The peer device could be
another RTP endpoint, or it could be an RTP middlebox that
redistributes the RTP packet streams to other RTP endpoints. This
limitation means that some of the RTP middlebox-based topologies are
not suitable for use in the WebRTC environment. Specifically: Video switching MCUs (Topo-Video-switch-MCU) SHOULD NOT be
used, since they make the use of RTCP for congestion control and
quality of service reports problematic (see Section 3.8 of ).The Relay-Transport Translator (Topo-PtM-Trn-Translator)
topology SHOULD NOT be used because its safe use requires a
congestion control algorithm or RTP circuit breaker that handles
point to multipoint, which has not yet been standardised.The following topology can be used, however it has some issues
worth noting:Content modifying MCUs with RTCP termination
(Topo-RTCP-terminating-MCU) MAY be used. Note that in this RTP
Topology, RTP loop detection and identification of active senders
is the responsibility of the WebRTC application; since the clients
are isolated from each other at the RTP layer, RTP cannot assist
with these functions (see section 3.9 of ).The RTP extensions described in to are designed to be used with centralised
conferencing, where an RTP middlebox (e.g., a conference bridge)
receives a participant's RTP packet streams and distributes them to
the other participants. These extensions are not necessary for
interoperability; an RTP end-point that does not implement these
extensions will work correctly, but might offer poor performance.
Support for the listed extensions will greatly improve the quality of
experience and, to provide a reasonable baseline quality, some of
these extensions are mandatory to be supported by WebRTC
end-points.The RTCP conferencing extensions are defined in Extended RTP Profile for Real-time Transport Control
Protocol (RTCP)-Based Feedback (RTP/AVPF) and the memo on Codec Control Messages (CCM) in RTP/AVPF; they
are fully usable by the Secure variant of this
profile (RTP/SAVPF).The Full Intra Request message is defined in Sections 3.5.1 and
4.3.1 of the Codec Control Messages.
It is used to make the mixer request a new Intra picture from a
participant in the session. This is used when switching between
sources to ensure that the receivers can decode the video or other
predictive media encoding with long prediction chains. WebRTC
senders MUST understand and react to FIR feedback messages they
receiver, since this greatly improves the user experience when using
centralised mixer-based conferencing. Support for sending FIR
messages is OPTIONAL.The Picture Loss Indication message is defined in Section 6.3.1
of the RTP/AVPF profile. It is used by
a receiver to tell the sending encoder that it lost the decoder
context and would like to have it repaired somehow. This is
semantically different from the Full Intra Request above as there
could be multiple ways to fulfil the request. WebRTC senders MUST
understand and react to PLI feedback messages as a loss tolerance
mechanism. Receivers MAY send PLI messages.The Slice Loss Indication message is defined in Section 6.3.2 of
the RTP/AVPF profile. It is used by a
receiver to tell the encoder that it has detected the loss or
corruption of one or more consecutive macro blocks, and would like
to have these repaired somehow. It is RECOMMENDED that receivers
generate SLI feedback messages if slices are lost when using a codec
that supports the concept of macro blocks. A sender that receives an
SLI feedback message SHOULD attempt to repair the lost slice(s).Reference Picture Selection Indication (RPSI) messages are
defined in Section 6.3.3 of the RTP/AVPF
profile . Some video encoding standards allow the use of
older reference pictures than the most recent one for predictive
coding. If such a codec is in use, and if the encoder has learnt
that encoder-decoder synchronisation has been lost, then a known as
correct reference picture can be used as a base for future coding.
The RPSI message allows this to be signalled. Receivers that detect
that encoder-decoder synchronisation has been lost SHOULD generate
an RPSI feedback message if codec being used supports reference
picture selection. A RTP packet stream sender that receives such an
RPSI message SHOULD act on that messages to change the reference
picture, if it is possible to do so within the available bandwidth
constraints, and with the codec being used.The temporal-spatial trade-off request and notification are
defined in Sections 3.5.2 and 4.3.2 of .
This request can be used to ask the video encoder to change the
trade-off it makes between temporal and spatial resolution, for
example to prefer high spatial image quality but low frame rate.
Support for TSTR requests and notifications is OPTIONAL.The TMMBR feedback message is defined in Sections 3.5.4 and 4.2.1
of the Codec Control Messages. This
request and its notification message are used by a media receiver to
inform the sending party that there is a current limitation on the
amount of bandwidth available to this receiver. This can be various
reasons for this: for example, an RTP mixer can use this message to
limit the media rate of the sender being forwarded by the mixer
(without doing media transcoding) to fit the bottlenecks existing
towards the other session participants. WebRTC senders are REQUIRED
to implement support for TMMBR messages, and MUST follow bandwidth
limitations set by a TMMBR message received for their SSRC. The
sending of TMMBR requests is OPTIONAL.The RTP specification provides the
capability to include RTP header extensions containing in-band data,
but the format and semantics of the extensions are poorly specified.
The use of header extensions is OPTIONAL in the WebRTC context, but if
they are used, they MUST be formatted and signalled following the
general mechanism for RTP header extensions defined in , since this gives well-defined semantics to RTP
header extensions.As noted in , the requirement from the RTP
specification that header extensions are "designed so that the header
extension may be ignored" stands. To be
specific, header extensions MUST only be used for data that can safely
be ignored by the recipient without affecting interoperability, and
MUST NOT be used when the presence of the extension has changed the
form or nature of the rest of the packet in a way that is not
compatible with the way the stream is signalled (e.g., as defined by
the payload type). Valid examples of RTP header extensions might
include metadata that is additional to the usual RTP information, but
that can safely be ignored without compromising interoperability.Many RTP sessions require synchronisation between audio, video,
and other content. This synchronisation is performed by receivers,
using information contained in RTCP SR packets, as described in the
RTP specification. This basic
mechanism can be slow, however, so it is RECOMMENDED that the rapid
RTP synchronisation extensions described in
be implemented in addition to RTCP SR-based synchronisation. The
rapid synchronisation extensions use the general RTP header
extension mechanism , which requires
signalling, but are otherwise backwards compatible.The Client to Mixer Audio Level
extension is an RTP header extension used by an endpoint to
inform a mixer about the level of audio activity in the packet to
which the header is attached. This enables an RTP middlebox to make
mixing or selection decisions without decoding or detailed
inspection of the payload, reducing the complexity in some types of
mixer. It can also save decoding resources in receivers, which can
choose to decode only the most relevant RTP packet streams based on
audio activity levels.The Client-to-Mixer Audio Level
header extension is RECOMMENDED to be implemented. If this header
extension is implemented, it is REQUIRED that implementations are
capable of encrypting the header extension according to since the information contained in these header
extensions can be considered sensitive. It is further RECOMMENDED
that this encryption is used, unless the encryption has been
explicitly disabled through API or signalling.The Mixer to Client Audio Level header
extension provides an endpoint with the audio level of the
different sources mixed into a common mix by a RTP mixer. This
enables a user interface to indicate the relative activity level of
each session participant, rather than just being included or not
based on the CSRC field. This is a pure optimisations of non
critical functions, and is hence OPTIONAL to implement. If this
header extension is implemented, it is REQUIRED that implementations
are capable of encrypting the header extension according to since the information contained in these header
extensions can be considered sensitive. It is further RECOMMENDED
that this encryption is used, unless the encryption has been
explicitly disabled through API or signalling.There are tools that can make RTP packet streams robust against
packet loss and reduce the impact of loss on media quality. However,
they all add overhead compared to a non-robust stream. The overhead
needs to be considered, and the aggregate bit-rate MUST be rate
controlled to avoid causing network congestion (see ). As a result, improving robustness might
require a lower base encoding quality, but has the potential to deliver
that quality with fewer errors. The mechanisms described in the
following sub-sections can be used to improve tolerance to packet
loss.As a consequence of supporting the RTP/SAVPF profile,
implementations can send negative acknowledgements (NACKs) for RTP
data packets . This feedback can be used to
inform a sender of the loss of particular RTP packets, subject to the
capacity limitations of the RTCP feedback channel. A sender can use
this information to optimise the user experience by adapting the media
encoding to compensate for known lost packets.RTP packet stream Senders are REQUIRED to understand the Generic
NACK message defined in Section 6.2.1 of , but
MAY choose to ignore some or all of this feedback (following Section
4.2 of ). Receivers MAY send NACKs for missing
RTP packets. Guidelines on when to send NACKs are provided in . It is not expected that a receiver will send a
NACK for every lost RTP packet, rather it needs to consider the cost
of sending NACK feedback, and the importance of the lost packet, to
make an informed decision on whether it is worth telling the sender
about a packet loss event.The RTP Retransmission Payload Format
offers the ability to retransmit lost packets based on NACK feedback.
Retransmission needs to be used with care in interactive real-time
applications to ensure that the retransmitted packet arrives in time
to be useful, but can be effective in environments with relatively low
network RTT (an RTP sender can estimate the RTT to the receivers using
the information in RTCP SR and RR packets, as described at the end of
Section 6.4.1 of ). The use of retransmissions
can also increase the forward RTP bandwidth, and can potentially
caused increased packet loss if the original packet loss was caused by
network congestion. We note, however, that retransmission of an
important lost packet to repair decoder state can have lower cost than
sending a full intra frame. It is not appropriate to blindly
retransmit RTP packets in response to a NACK. The importance of lost
packets and the likelihood of them arriving in time to be useful needs
to be considered before RTP retransmission is used.Receivers are REQUIRED to implement support for RTP retransmission
packets . Senders MAY send RTP retransmission
packets in response to NACKs if the RTP retransmission payload format
has been negotiated for the session, and if the sender believes it is
useful to send a retransmission of the packet(s) referenced in the
NACK. An RTP sender does not need to retransmit every NACKed
packet.The use of Forward Error Correction (FEC) can provide an effective
protection against some degree of packet loss, at the cost of steady
bandwidth overhead. There are several FEC schemes that are defined for
use with RTP. Some of these schemes are specific to a particular RTP
payload format, others operate across RTP packets and can be used with
any payload format. It needs to be noted that using redundant encoding
or FEC will lead to increased play out delay, which needs to be
considered when choosing the redundancy or FEC formats and their
respective parameters.If an RTP payload format negotiated for use in a RTCPeerConnection
supports redundant transmission or FEC as a standard feature of that
payload format, then that support MAY be used in the
RTCPeerConnection, subject to any appropriate signalling.There are several block-based FEC schemes that are designed for use
with RTP independent of the chosen RTP payload format. At the time of
this writing there is no consensus on which, if any, of these FEC
schemes is appropriate for use in the WebRTC context. Accordingly,
this memo makes no recommendation on the choice of block-based FEC for
WebRTC use.WebRTC will be used in heterogeneous network environments using a
variety set of link technologies, including both wired and wireless
links, to interconnect potentially large groups of users around the
world. As a result, the network paths between users can have widely
varying one-way delays, available bit-rates, load levels, and traffic
mixtures. Individual end-points can send one or more RTP packet streams
to each participant in a WebRTC conference, and there can be several
participants. Each of these RTP packet streams can contain different
types of media, and the type of media, bit rate, and number of RTP
packet streams as well as transport-layer flows can be highly
asymmetric. Non-RTP traffic can share the network paths with RTP
transport-layer flows. Since the network environment is not predictable
or stable, WebRTC end-points MUST ensure that the RTP traffic they
generate can adapt to match changes in the available network
capacity.The quality of experience for users of WebRTC implementation is very
dependent on effective adaptation of the media to the limitations of the
network. End-points have to be designed so they do not transmit
significantly more data than the network path can support, except for
very short time periods, otherwise high levels of network packet loss or
delay spikes will occur, causing media quality degradation. The limiting
factor on the capacity of the network path might be the link bandwidth,
or it might be competition with other traffic on the link (this can be
non-WebRTC traffic, traffic due to other WebRTC flows, or even
competition with other WebRTC flows in the same session).An effective media congestion control algorithm is therefore an
essential part of the WebRTC framework. However, at the time of this
writing, there is no standard congestion control algorithm that can be
used for interactive media applications such as WebRTC's flows. Some
requirements for congestion control algorithms for RTCPeerConnections
are discussed in . It is
expected that a future version of this memo will mandate the use of a
congestion control algorithm that satisfies these requirements.In the absence of a concrete congestion control algorithm, all
WebRTC implementations MUST implement the RTP circuit breaker
algorithm that is described in . The RTP circuit
breaker is designed to enable applications to recognise and react to
situations of extreme network congestion. However, since the RTP
circuit breaker might not be triggered until congestion becomes
extreme, it cannot be considered a substitute for congestion control,
and applications MUST also implement congestion control to allow them
to adapt to changes in network capacity. Any future RTP congestion
control algorithms are expected to operate within the envelope allowed
by the circuit breaker.The session establishment signalling will also necessarily
establish boundaries to which the media bit-rate will conform. The
choice of media codecs provides upper- and lower-bounds on the
supported bit-rates that the application can utilise to provide useful
quality, and the packetization choices that exist. In addition, the
signalling channel can establish maximum media bit-rate boundaries
using the SDP "b=AS:" or "b=CT:" lines, and the RTP/AVPF Temporary
Maximum Media Stream Bit Rate (TMMBR) Requests (see of this memo). The combination of media codec
choice and signalled bandwidth limits SHOULD be used to limit traffic
based on known bandwidth limitations, for example the capacity of the
edge links, to the extent possible.Experience with the congestion control algorithms of TCP , TFRC , and DCCP , , , has shown that feedback on packet arrivals needs
to be sent frequently (roughly once per round trip time is common). We
note that the real-time media traffic might not be able to adapt to
changing path conditions as rapidly as elastic applications using TCP,
but frequent feedback, perhaps on the order of once per video frame,
is still needed to allow the congestion control algorithm to track the
path dynamics.As an example of the type of RTCP congestion control feedback that
is possible, consider one of the simplest scenarios for WebRTC: a
point to point video call between two end systems. There will be four
RTP flows in this scenario, two audio and two video, with all four
flows being active for essentially all the time (the audio flows will
likely use voice activity detection and comfort noise to reduce the
packet rate during silent periods, but doesn't cause transmissions to
stop). Assume all four flows are sent in a single RTP session, each
using a separate SSRC. Further, assume each SSRC sends RTCP reports
for all other SSRCs in the session (i.e., the optimisations in are not
used, giving the worst case for the RTCP overhead). When all members
are senders like this, the RTCP timing rules in Sections 6.2 and 6.3
of and reduce
to:where avg_rtcp_size is measured in octets, and the rtcp_bw is the
bandwidth available for RTCP. The average RTCP size will depend on the
amount of feedback that is sent in each RTCP packet, on the number of
members in the session, and on the size of source description (RTCP
SDES) information sent. As a baseline, each RTCP packet will be a
compound RTCP packet that contains an RTCP SR and an RTCP SDES packet.
In the scenario above, each RTCP SR packet will contain three report
blocks, once for each of the other RTP SSRCs sending data, for a total
of 100 octets (this is 8 octets header, 20 octets sender info, and 3 *
24 octets report blocks). The RTCP SDES packet will comprise a header
(4 octets), an originating SSRC (4 octets), a CNAME chunk, and
padding. If the CNAME follows and it will be
19 octets in size, and require 1 octet of padding. The resulting
compound RTCP packet will be 128 octets in size. If sent in UDP/IPv4
with no IP options and using Secure RTP, which adds 20 (IPv4) + 8
(UDP) + 14 (SRTP with 80 bit Authentication tag), the avg_rtcp_size
will therefore be 170 octets, including the header overhead. The value
n is this scenario is 4, and the rtcp_bw is assumed to be 5% of the
session bandwidth.If it is desired to send RTCP feedback packets on average 30 times
per second, to correspond to one RTCP report every frame for 30fps
video, we can invert the above rtcp_interval calculation to get an
rtcp_bw that gives an interval of 1/30th of a second or lower. This
corresponds to an rtcp_bw of 20400 octets per second (since 1/30 = 170
* 4 / 20400). This is 163200 bits per second, which if 5% of the
session bandwidth, gives a session bandwidth of approximately 3.3Mbps
(i.e., 3.3Mbps media rate, plus an additional 5% for RTCP, to give a
total data rate of approximately 3.4Mbps). That is, RTCP can report on
every frame of video provided the session bandwidth is 3.3Mbps or
larger, when every SSRC sends a report for every video frame. Please
note that the actual RTCP transmission intervals will be within the
interval [0.0135, 0.0406]s, but maintaining an average RTCP
transmission interval of 0.033s.Note: To achieve the RTCP transmission intervals above the
RTP/SAVPF profile with T_rr_interval=0 is used, since even when
using the reduced minimal transmission interval, the RTP/SAVP
profile would only allow sending RTCP at most every 0.11s (every
third frame of video). Using RTP/SAVPF with T_rr_interval=0
however is capable of fully utilizing the configured 5% RTCP
bandwidth fraction.If additional feedback beyond the standard report block is needed,
the session bandwidth needed will increase. For example, with an
additional 20 octets data being reported in each RTCP packet, the
session bandwidth needed increases to 3.5Mbps for every SSRC to be
able to report on every frame. However, the above baseline might not
be the most appropriate usage of the RTCP bandwidth. Depending on
needs, a less frequent usage of regular RTCP compound packets,
controlled by T_rr_interval combined with using the reduced size RTCP
packets, can achieve more frequent and useful reporting. Also the
reporting requirements defined in will reduced
the amount of bandwidth consumed for reporting when each endpoint has
multiple SSRCs.Calculations such as these show that RTCP cannot be used to send
per-packet congestion feedback. RTCP can, however, be used to send
congestion feedback on each frame of video sent in an interactive
video conferencing scenario, provided the RTCP parameters are
correctly configured and the overall session bandwidth exceeds a
couple of megabits per second (the exact rate depending on the number
of session participants, the RTCP bandwidth fraction, and whether
audio and video are sent in one or two RTP sessions). Using similar
calculations, it can be shown that RTCP can likely also be used to
send feedback on a per-RTT basis, provided the RTT is not too low.Interactive communication might not be able to afford to wait for
packet losses to occur to indicate congestion, because an increase in
play out delay due to queuing (most prominent in wireless networks)
can easily lead to packets being dropped due to late arrival at the
receiver. Therefore, more sophisticated cues might need to be reported
-- to be defined in a suitable congestion control framework as noted
above -- which, in turn, increase the report size again. For example,
different RTCP XR report blocks (jointly) provide the necessary
details to implement a variety of congestion control algorithms, but
the (compound) report size grows quickly.There are legacy RTP implementations that do not implement RTCP,
and hence do not provide any congestion feedback. Congestion control
cannot be performed with these end-points. WebRTC implementations that
need to interwork with such end-points MUST limit their transmission
to a low rate, equivalent to a VoIP call using a low bandwidth codec,
that is unlikely to cause any significant congestion.When interworking with legacy implementations that support RTCP
using the RTP/AVP profile, congestion
feedback is provided in RTCP RR packets every few seconds.
Implementations that have to interwork with such end-points MUST
ensure that they keep within the RTP circuit
breaker constraints to limit the congestion they can cause.If a legacy end-point supports RTP/AVPF, this enables negotiation
of important parameters for frequent reporting, such as the "trr-int"
parameter, and the possibility that the end-point supports some useful
feedback format for congestion control purpose such as TMMBR. Implementations that have to interwork
with such end-points MUST ensure that they stay within the RTP circuit
breaker constraints to limit the congestion they can cause, but
might find that they can achieve better congestion response depending
on the amount of feedback that is available.With proprietary congestion control algorithms issues can arise
when different algorithms and implementations interact in a
communication session. If the different implementations have made
different choices in regards to the type of adaptation, for example
one sender based, and one receiver based, then one could end up in
situation where one direction is dual controlled, when the other
direction is not controlled. This memo cannot mandate behaviour for
proprietary congestion control algorithms, but implementations that
use such algorithms ought to be aware of this issue, and try to ensure
that both effective congestion control is negotiated for media flowing
in both directions. If the IETF were to standardise both sender- and
receiver-based congestion control algorithms for WebRTC traffic in the
future, the issues of interoperability, control, and ensuring that
both directions of media flow are congestion controlled would also
need to be considered.As described in , implementations are
REQUIRED to generate RTCP Sender Report (SR) and Reception Report (RR)
packets relating to the RTP packet streams they send and receive. These
RTCP reports can be used for performance monitoring purposes, since they
include basic packet loss and jitter statistics.A large number of additional performance metrics are supported by the
RTCP Extended Reports (XR) framework . At the time of this writing, it is not clear what
extended metrics are suitable for use in the WebRTC context, so there is
no requirement that implementations generate RTCP XR packets. However,
implementations that can use detailed performance monitoring data MAY
generate RTCP XR packets as appropriate; the use of such packets SHOULD
be signalled in advance.All WebRTC implementations MUST be prepared to receive RTP XR report
packets, whether or not they were signalled. There is no requirement
that the data contained in such reports be used, or exposed to the
Javascript application, however.It is possible that the core set of RTP protocols and RTP extensions
specified in this memo will prove insufficient for the future needs of
WebRTC applications. In this case, future updates to this memo MUST be
made following the Guidelines for Writers of RTP
Payload Format Specifications , How to Write an RTP Payload
Format and Guidelines for Extending the
RTP Control Protocol, and SHOULD take into account any future
guidelines for extending RTP and related protocols that have been
developed.Authors of future extensions are urged to consider the wide range of
environments in which RTP is used when recommending extensions, since
extensions that are applicable in some scenarios can be problematic in
others. Where possible, the WebRTC framework will adopt RTP extensions
that are of general utility, to enable easy implementation of a gateway
to other applications using RTP, rather than adopt mechanisms that are
narrowly targeted at specific WebRTC use cases.RTP is built with the assumption that an external signalling channel
exists, and can be used to configure RTP sessions and their features.
The basic configuration of an RTP session consists of the following
parameters:The name of the RTP profile to be used in
session. The RTP/AVP and RTP/AVPF profiles can interoperate on basic
level, as can their secure variants RTP/SAVP and RTP/SAVPF. The secure variants of the
profiles do not directly interoperate with the non-secure variants,
due to the presence of additional header fields for authentication
in SRTP packets and cryptographic transformation of the payload.
WebRTC requires the use of the RTP/SAVPF profile, and this MUST be
signalled if SDP is used. Interworking functions might transform
this into the RTP/SAVP profile for a legacy use case, by indicating
to the WebRTC end-point that the RTP/SAVPF is used, and limiting the
usage of the "a=rtcp-fb:" attribute to indicate a trr-int value of 4
seconds.Source and destination IP
address(s) and ports for RTP and RTCP MUST be signalled for each RTP
session. In WebRTC these transport addresses will be provided by ICE
that signals candidates and arrives at nominated candidate address
pairs. If RTP and RTCP multiplexing is
to be used, such that a single port, i.e. transport-layer flow, is
used for RTP and RTCP flows, this MUST be signalled (see ).The
mapping between media type names (and hence the RTP payload formats
to be used), and the RTP payload type numbers MUST be signalled.
Each media type MAY also have a number of media type parameters that
MUST also be signalled to configure the codec and RTP payload format
(the "a=fmtp:" line from SDP). of this
memo discusses requirements for uniqueness of payload types.The RTP extensions to be used SHOULD
be agreed upon, including any parameters for each respective
extension. At the very least, this will help avoiding using
bandwidth for features that the other end-point will ignore. But for
certain mechanisms there is requirement for this to happen as
interoperability failure otherwise happens.Support for exchanging RTCP Bandwidth
values to the end-points will be necessary. This SHALL be done as
described in "Session Description Protocol
(SDP) Bandwidth Modifiers for RTP Control Protocol (RTCP)
Bandwidth", or something semantically equivalent. This also
ensures that the end-points have a common view of the RTCP
bandwidth, this is important as too different view of the bandwidths
can lead to failure to interoperate.These parameters are often expressed in SDP messages conveyed within
an offer/answer exchange. RTP does not depend on SDP or on the
offer/answer model, but does require all the necessary parameters to be
agreed upon, and provided to the RTP implementation. We note that in the
WebRTC context it will depend on the signalling model and API how these
parameters need to be configured but they will be need to either set in
the API or explicitly signalled between the peers.The WebRTC API and the
Media Capture and
Streams API defines and uses the concept of a MediaStream that
consists of zero or more MediaStreamTracks. A MediaStreamTrack is an
individual stream of media from any type of media source like a
microphone or a camera, but also conceptual sources, like a audio mix or
a video composition, are possible. The MediaStreamTracks within a
MediaStream need to be possible to play out synchronised.A MediaStreamTrack's realisation in RTP in the context of an
RTCPeerConnection consists of a source packet stream identified with an
SSRC within an RTP session part of the RTCPeerConnection. The
MediaStreamTrack can also result in additional packet streams, and thus
SSRCs, in the same RTP session. These can be dependent packet streams
from scalable encoding of the source stream associated with the
MediaStreamTrack, if such a media encoder is used. They can also be
redundancy packet streams, these are created when applying Forward Error Correction or RTP retransmission to the source packet
stream.It is important to note that the same media source can be feeding
multiple MediaStreamTracks. As different sets of constraints or other
parameters can be applied to the MediaStreamTrack, each MediaStreamTrack
instance added to a RTCPeerConnection SHALL result in an independent
source packet stream, with its own set of associated packet streams, and
thus different SSRC(s). It will depend on applied constraints and
parameters if the source stream and the encoding configuration will be
identical between different MediaStreamTracks sharing the same media
source. Thus it is possible for multiple source packet streams to share
encoded streams (but not packet streams), but this is an implementation
choice to try to utilise such optimisations. Note that such
optimizations would need to take into account that the constraints for
one of the MediaStreamTracks can at any moment change, meaning that the
encoding configurations might no longer be identical.The same MediaStreamTrack can also be included in multiple
MediaStreams, thus multiple sets of MediaStreams can implicitly need to
use the same synchronisation base. To ensure that this works in all
cases, and don't forces a end-point to change synchronisation base and
CNAME in the middle of a ongoing delivery of any packet streams, which
would cause media disruption; all MediaStreamTracks and their associated
SSRCs originating from the same end-point needs to be sent using the
same CNAME within one RTCPeerConnection. This is motivating the strong
recommendation in to only use a single CNAME.
The requirement on using the same CNAME for all SSRCs that
originates from the same end-point, does not require middleboxes
that forwards traffic from multiple end-points to only use a single
CNAME.Different CNAMEs normally need to be used for different
RTCPeerConnection instances, as specified in .
Having two communication sessions with the same CNAME could enable
tracking of a user or device across different services (see Section
4.4.1 of for details). A web
application can request that the CNAMEs used in different
RTCPeerConnection within a same-orign context to be the same, this allow
for synchronization of the endpoint's RTP packet streams across the
different RTCPeerConnections.Note: this doesn't result in a tracking issue, since the creation
of matching CNAMEs depends on existing tracking.The above will currently force a WebRTC end-point that receives
an MediaStreamTrack on one RTCPeerConnection and adds it as an outgoing
on any RTCPeerConnection to perform resynchronisation of the stream.
This, as the sending party needs to change the CNAME, which implies that
it has to use a locally available system clock as timebase for the
synchronisation. Thus, the relative relation between the timebase of the
incoming stream and the system sending out needs to defined. This
relation also needs monitoring for clock drift and likely adjustments of
the synchronisation. The sending entity is also responsible for
congestion control for its the sent streams. In cases of packet loss the
loss of incoming data also needs to be handled. This leads to the
observation that the method that is least likely to cause issues or
interruptions in the outgoing source packet stream is a model of full
decoding, including repair etc followed by encoding of the media again
into the outgoing packet stream. Optimisations of this method is clearly
possible and implementation specific.A WebRTC end-point MUST support receiving multiple MediaStreamTracks,
where each of different MediaStreamTracks (and their sets of associated
packet streams) uses different CNAMEs. However, MediaStreamTracks that
are received with different CNAMEs have no defined synchronisation.Note: The motivation for supporting reception of multiple CNAMEs
are to allow for forward compatibility with any future changes that
enables more efficient stream handling when end-points relay/forward
streams. It also ensures that end-points can interoperate with
certain types of multi-stream middleboxes or end-points that are not
WebRTC.The binding between the WebRTC MediaStreams, MediaStreamTracks and
the SSRC is done as specified in "Cross Session Stream Identification in
the Session Description Protocol". This document also defines, in
section 4.1, how to map unknown source packet stream SSRCs to
MediaStreamTracks and MediaStreams. Commonly the RTP Payload Type of any
incoming packets will reveal if the packet stream is a source stream or
a redundancy or dependent packet stream. The association to the correct
source packet stream depends on the payload format in use for the packet
stream.Finally this specification puts a requirement on the WebRTC API to
realize a method for determining the CSRC
list as well as the Mixer-to-Client audio levels (when
supported) and the basic requirements for this is further discussed in
.The following discussion provides some guidance on the implementation
of the RTP features described in this memo. The focus is on a WebRTC
end-point implementation perspective, and while some mention is made of
the behaviour of middleboxes, that is not the focus of this memo.A WebRTC end-point will be a simultaneous participant in one or
more RTP sessions. Each RTP session can convey multiple media sources,
and can include media data from multiple end-points. In the following,
we outline some ways in which WebRTC end-points can configure and use
RTP sessions.RTP is a group communication protocol, and every RTP session can
potentially contain multiple RTP packet streams. There are several
reasons why this might be desirable: Outside of WebRTC, it is
common to use one RTP session for each type of media sources
(e.g., one RTP session for audio sources and one for video
sources, each sent over different transport layer flows).
However, to reduce the number of UDP ports used, the default in
WebRTC is to send all types of media in a single RTP session, as
described in , using RTP and
RTCP multiplexing () to further
reduce the number of UDP ports needed. This RTP session then
uses only one bi-directional transport-layer flow, but will
contain multiple RTP packet streams, each containing a different
type of media. A common example might be an end-point with a
camera and microphone that sends two RTP packet streams, one
video and one audio, into a single RTP session.A WebRTC end-point might
have multiple cameras, microphones, or other media capture
devices, and so might want to generate several RTP packet
streams of the same media type. Alternatively, it might want to
send media from a single capture device in several different
formats or quality settings at once. Both can result in a single
end-point sending multiple RTP packet streams of the same media
type into a single RTP session at the same time.An end-point might send a
RTP packet stream that is somehow associated with another
stream. For example, it might send an RTP packet stream that
contains FEC or retransmission data relating to another stream.
Some RTP payload formats send this sort of associated repair
data as part of the source packet stream, while others send it
as a separate packet stream.An
end-point can use a layered media codec, for example H.264 SVC,
or a multiple description codec, that generates multiple RTP
packet streams, each with a distinct RTP SSRC, within a single
RTP session.An
RTP session, in the WebRTC context, is a point-to-point
association between an end-point and some other peer device,
where those devices share a common SSRC space. The peer device
might be another WebRTC end-point, or it might be an RTP mixer,
translator, or some other form of media processing middlebox. In
the latter cases, the middlebox might send mixed or relayed RTP
streams from several participants, that the WebRTC end-point
will need to render. Thus, even though a WebRTC end-point might
only be a member of a single RTP session, the peer device might
be extending that RTP session to incorporate other end-points.
WebRTC is a group communication environment and end-points need
to be capable of receiving, decoding, and playing out multiple
RTP packet streams at once, even in a single RTP session.In addition to sending and receiving multiple RTP packet streams
within a single RTP session, a WebRTC end-point might participate in
multiple RTP sessions. There are several reasons why a WebRTC
end-point might choose to do this: The common
practice in the non-WebRTC world is to send different types of
media in separate RTP sessions, for example using one RTP
session for audio and another RTP session, on a separate
transport layer flow, for video. All WebRTC end-points need to
support the option of sending different types of media on
different RTP sessions, so they can interwork with such legacy
devices. This is discussed further in .Some
network-based quality of service mechanisms operate on the
granularity of transport layer flows. If it is desired to use
these mechanisms to provide differentiated quality of service
for some RTP packet streams, then those RTP packet streams need
to be sent in a separate RTP session using a different
transport-layer flow, and with appropriate quality of service
marking. This is discussed further in .An
end-point might want to send RTP packet streams that have
different purposes on different RTP sessions, to make it easy
for the peer device to distinguish them. For example, some
centralised multiparty conferencing systems display the active
speaker in high resolution, but show low resolution "thumbnails"
of other participants. Such systems might configure the
end-points to send simulcast high- and low-resolution versions
of their video using separate RTP sessions, to simplify the
operation of the RTP middlebox. In the WebRTC context this is
currently possible to accomplished by establishing multiple
WebRTC MediaStreamTracks that have the same media source in one
(or more) RTCPeerConnection. Each MediaStreamTrack is then
configured to deliver a particular media quality and thus media
bit-rate, and will produce an independently encoded version with
the codec parameters agreed specifically in the context of that
RTCPeerConnection. The RTP middlebox can distinguish packets
corresponding to the low- and high-resolution streams by
inspecting their SSRC, RTP payload type, or some other
information contained in RTP payload, RTP header extension or
RTCP packets, but it can be easier to distinguish the RTP packet
streams if they arrive on separate RTP sessions on separate
transport-layer flows.A
multi-party conference does not need to use an RTP middlebox.
Rather, a multi-unicast mesh can be created, comprising several
distinct RTP sessions, with each participant sending RTP traffic
over a separate RTP session (that is, using an independent
RTCPeerConnection object) to every other participant, as shown
in . This topology has the benefit of
not requiring an RTP middlebox node that is trusted to access
and manipulate the media data. The downside is that it increases
the used bandwidth at each sender by requiring one copy of the
RTP packet streams for each participant that are part of the
same session beyond the sender itself.The multi-unicast topology could also be implemented as a
single RTP session, spanning multiple peer-to-peer transport
layer connections, or as several pairwise RTP sessions, one
between each pair of peers. To maintain a coherent mapping
between the relation between RTP sessions and RTCPeerConnection
objects we recommend that this is implemented as several
individual RTP sessions. The only downside is that end-point A
will not learn of the quality of any transmission happening
between B and C, since it will not see RTCP reports for the RTP
session between B and C, whereas it would it all three
participants were part of a single RTP session. Experience with
the Mbone tools (experimental RTP-based multicast conferencing
tools from the late 1990s) has showed that RTCP reception
quality reports for third parties can usefully be presented to
the users in a way that helps them understand asymmetric network
problems, and the approach of using separate RTP sessions
prevents this. However, an advantage of using separate RTP
sessions is that it enables using different media bit-rates and
RTP session configurations between the different peers, thus not
forcing B to endure the same quality reductions if there are
limitations in the transport from A to C as C will. It it
believed that these advantages outweigh the limitations in
debugging power.A
common scenario in multi-party conferencing is to create
indirect connections to multiple peers, using an RTP mixer,
translator, or some other type of RTP middlebox. outlines a simple topology that might
be used in a four-person centralised conference. The middlebox
acts to optimise the transmission of RTP packet streams from
certain perspectives, either by only sending some of the
received RTP packet stream to any given receiver, or by
providing a combined RTP packet stream out of a set of
contributing streams.There are various methods of implementation for the
middlebox. If implemented as a standard RTP mixer or translator,
a single RTP session will extend across the middlebox and
encompass all the end-points in one multi-party session. Other
types of middlebox might use separate RTP sessions between each
end-point and the middlebox. A common aspect is that these RTP
middleboxes can use a number of tools to control the media
encoding provided by a WebRTC end-point. This includes functions
like requesting breaking the encoding chain and have the encoder
produce a so called Intra frame. Another is limiting the
bit-rate of a given stream to better suit the mixer view of the
multiple down-streams. Others are controlling the most suitable
frame-rate, picture resolution, the trade-off between frame-rate
and spatial quality. The middlebox gets the significant
responsibility to correctly perform congestion control, source
identification, manage synchronisation while providing the
application with suitable media optimizations. The middlebox is
also has to be a trusted node when it comes to security, since
it manipulates either the RTP header or the media itself (or
both) received from one end-point, before sending it on towards
the end-point(s), thus they need to be able to decrypt and then
encrypt it before sending it out.RTP Mixers can create a situation where an end-point
experiences a situation in-between a session with only two
end-points and multiple RTP sessions. Mixers are expected to not
forward RTCP reports regarding RTP packet streams across
themselves. This is due to the difference in the RTP packet
streams provided to the different end-points. The original media
source lacks information about a mixer's manipulations prior to
sending it the different receivers. This scenario also results
in that an end-point's feedback or requests goes to the mixer.
When the mixer can't act on this by itself, it is forced to go
to the original media source to fulfil the receivers request.
This will not necessarily be explicitly visible any RTP and RTCP
traffic, but the interactions and the time to complete them will
indicate such dependencies.Providing source authentication in multi-party scenarios is a
challenge. In the mixer-based topologies, end-points source
authentication is based on, firstly, verifying that media comes
from the mixer by cryptographic verification and, secondly,
trust in the mixer to correctly identify any source towards the
end-point. In RTP sessions where multiple end-points are
directly visible to an end-point, all end-points will have
knowledge about each others' master keys, and can thus inject
packets claimed to come from another end-point in the session.
Any node performing relay can perform non-cryptographic
mitigation by preventing forwarding of packets that have SSRC
fields that came from other end-points before. For cryptographic
verification of the source SRTP would require additional
security mechanisms, for example TESLA
for SRTP, that are not part of the base WebRTC
standards.It is
sometimes desirable for an end-point that receives an RTP packet
stream to be able to forward that RTP packet stream to a third
party. The are some obvious security and privacy implications in
supporting this, but also potential uses. This is supported in
the W3C API by taking the received and decoded media and using
it as media source that is re-encoding and transmitted as a new
stream.At the RTP layer, media forwarding acts as a back-to-back RTP
receiver and RTP sender. The receiving side terminates the RTP
session and decodes the media, while the sender side re-encodes
and transmits the media using an entirely separate RTP session.
The original sender will only see a single receiver of the
media, and will not be able to tell that forwarding is happening
based on RTP-layer information since the RTP session that is
used to send the forwarded media is not connected to the RTP
session on which the media was received by the node doing the
forwarding.The end-point that is performing the forwarding is
responsible for producing an RTP packet stream suitable for
onwards transmission. The outgoing RTP session that is used to
send the forwarded media is entirely separate to the RTP session
on which the media was received. This will require media
transcoding for congestion control purpose to produce a suitable
bit-rate for the outgoing RTP session, reducing media quality
and forcing the forwarding end-point to spend the resource on
the transcoding. The media transcoding does result in a
separation of the two different legs removing almost all
dependencies, and allowing the forwarding end-point to optimize
its media transcoding operation. The cost is greatly increased
computational complexity on the forwarding node. Receivers of
the forwarded stream will see the forwarding device as the
sender of the stream, and will not be able to tell from the RTP
layer that they are receiving a forwarded stream rather than an
entirely new RTP packet stream generated by the forwarding
device.There are use cases for differentiated treatment of RTP packet
streams. Such differentiation can happen at several places in the
system. First of all is the prioritization within the end-point
sending the media, which controls, both which RTP packet streams
that will be sent, and their allocation of bit-rate out of the
current available aggregate as determined by the congestion
control.It is expected that the WebRTC API will allow the
application to indicate relative priorities for different
MediaStreamTracks. These priorities can then be used to influence
the local RTP processing, especially when it comes to congestion
control response in how to divide the available bandwidth between
the RTP packet streams. Any changes in relative priority will also
need to be considered for RTP packet streams that are associated
with the main RTP packet streams, such as redundant streams for RTP
retransmission and FEC. The importance of such redundant RTP packet
streams is dependent on the media type and codec used, in regards to
how robust that codec is to packet loss. However, a default policy
might to be to use the same priority for redundant RTP packet stream
as for the source RTP packet stream.Secondly, the network can prioritize transport-layer flows and
sub-flows, including RTP packet streams. Typically, differential
treatment includes two steps, the first being identifying whether an
IP packet belongs to a class that has to be treated differently, the
second the actual mechanism to prioritize packets. This is done
according to three methods: The end-point marks a packet with a
DiffServ code point to indicate to the network that the packet
belongs to a particular class.Packets that need to be given a
particular treatment are identified using a combination of IP
and port address.A network classifier (DPI)
inspects the packet and tries to determine if the packet
represents a particular application and type that is to be
prioritized.Flow-based differentiation will provide the same treatment to all
packets within a transport-layer flow, i.e., relative prioritization
is not possible. Moreover, if the resources are limited it might not
be possible to provide differential treatment compared to
best-effort for all the RTP packet streams in a WebRTC application.
When flow-based differentiation is available the WebRTC application
needs to know about it so that it can provide the separation of the
RTP packet streams onto different UDP flows to enable a more
granular usage of flow based differentiation. That way at least
providing different prioritization of audio and video if desired by
application.DiffServ assumes that either the end-point or a classifier can
mark the packets with an appropriate DSCP so that the packets are
treated according to that marking. If the end-point is to mark the
traffic two requirements arise in the WebRTC context: 1) The WebRTC
application or browser has to know which DSCP to use and that it can
use them on some set of RTP packet streams. 2) The information needs
to be propagated to the operating system when transmitting the
packet. Details of this process are outside the scope of this memo
and are further discussed in "DSCP and other packet markings
for RTCWeb QoS".For packet based marking schemes it might be possible to mark
individual RTP packets differently based on the relative priority of
the RTP payload. For example video codecs that have I, P, and B
pictures could prioritise any payloads carrying only B frames less,
as these are less damaging to loose. However, depending on the QoS
mechanism and what markings that are applied, this can result in not
only different packet drop probabilities but also packet reordering,
see for further
discussion. As default policy all RTP packets related to a RTP
packet stream ought to be provided with the same prioritization;
per-packet prioritization is outside the scope of this memo, but
might be specified elsewhere in future.It is also important to consider how RTCP packets associated with
a particular RTP packet stream need to be marked. RTCP compound
packets with Sender Reports (SR), ought to be marked with the same
priority as the RTP packet stream itself, so the RTCP-based
round-trip time (RTT) measurements are done using the same
transport-layer flow priority as the RTP packet stream experiences.
RTCP compound packets containing RR packet ought to be sent with the
priority used by the majority of the RTP packet streams reported on.
RTCP packets containing time-critical feedback packets can use
higher priority to improve the timeliness and likelihood of delivery
of such feedback.Each RTP packet stream is identified by a unique synchronisation
source (SSRC) identifier. The SSRC identifier is carried in each of
the RTP packets comprising a RTP packet stream, and is also used to
identify that stream in the corresponding RTCP reports. The SSRC is
chosen as discussed in . The first stage in
demultiplexing RTP and RTCP packets received on a single transport
layer flow at a WebRTC end-point is to separate the RTP packet
streams based on their SSRC value; once that is done, additional
demultiplexing steps can determine how and where to render the
media.RTP allows a mixer, or other RTP-layer middlebox, to combine
encoded streams from multiple media sources to form a new encoded
stream from a new media source (the mixer). The RTP packets in that
new RTP packet stream can include a Contributing Source (CSRC) list,
indicating which original SSRCs contributed to the combined source
stream. As described in ,
implementations need to support reception of RTP data packets
containing a CSRC list and RTCP packets that relate to sources
present in the CSRC list. The CSRC list can change on a
packet-by-packet basis, depending on the mixing operation being
performed. Knowledge of what media sources contributed to a
particular RTP packet can be important if the user interface
indicates which participants are active in the session. Changes in
the CSRC list included in packets needs to be exposed to the WebRTC
application using some API, if the application is to be able to
track changes in session participation. It is desirable to map CSRC
values back into WebRTC MediaStream identities as they cross this
API, to avoid exposing the SSRC/CSRC name space to JavaScript
applications.If the mixer-to-client audio level extension is being used in the session (see ), the information in the CSRC list is
augmented by audio level information for each contributing source.
This information can usefully be exposed in the user interface.The RTP standard requires any RTP
implementation to have support for detecting and handling SSRC
collisions, i.e., resolve the conflict when two different end-points
use the same SSRC value. This requirement also applies to WebRTC
end-points. There are several scenarios where SSRC collisions can
occur:In a point-to-point session where each SSRC is associated
with either of the two end-points and where the main media
carrying SSRC identifier will be announced in the signalling
channel, a collision is less likely to occur due to the
information about used SSRCs provided by Source-Specific SDP Attributes. Still,
collisions can occur if both end-points start uses an new SSRC
identifier prior to having signalled it to the peer and received
acknowledgement on the signalling message. The Source-Specific SDP Attributes contains
no mechanism to resolve SSRC collisions or reject a end-points
usage of an SSRC.SSRC values that have not been signalled could also appear in
an RTP session. This is more likely than it appears, since some
RTP functions use extra SSRCs to provide their functionality.
For example, retransmission data might be transmitted using a
separate RTP packet stream that requires its own SSRC, separate
to the SSRC of the source RTP packet stream . In those cases, an end-point can create a
new SSRC that strictly doesn't need to be announced over the
signalling channel to function correctly on both RTP and
RTCPeerConnection level.Multiple end-points in a multiparty conference can create new
sources and signal those towards the RTP middlebox. In cases
where the SSRC/CSRC are propagated between the different
end-points from the RTP middlebox collisions can occur.An RTP middlebox could connect an end-point's
RTCPeerConnection to another RTCPeerConnection from the same
end-point, thus forming a loop where the end-point will receive
its own traffic. While is is clearly considered a bug, it is
important that the end-point is able to recognise and handle the
case when it occurs. This case becomes even more problematic
when media mixers, and so on, are involved, where the stream
received is a different stream but still contains this client's
input.These SSRC/CSRC collisions can only be handled on RTP level as
long as the same RTP session is extended across multiple
RTCPeerConnections by a RTP middlebox. To resolve the more generic
case where multiple RTCPeerConnections are interconnected, then
identification of the media source(s) part of a MediaStreamTrack
being propagated across multiple interconnected RTCPeerConnection
needs to be preserved across these interconnections.When an end-point sends media from more than one media source, it
needs to consider if (and which of) these media sources are to be
synchronized. In RTP/RTCP, synchronisation is provided by having a
set of RTP packet streams be indicated as coming from the same
synchronisation context and logical end-point by using the same RTCP
CNAME identifier.The next provision is that the internal clocks of all media
sources, i.e., what drives the RTP timestamp, can be correlated to a
system clock that is provided in RTCP Sender Reports encoded in an
NTP format. By correlating all RTP timestamps to a common system
clock for all sources, the timing relation of the different RTP
packet streams, also across multiple RTP sessions can be derived at
the receiver and, if desired, the streams can be synchronized. The
requirement is for the media sender to provide the correlation
information; it is up to the receiver to use it or not.The overall security architecture for WebRTC is described in , and security considerations
for the WebRTC framework are described in . These considerations also apply to
this memo.The security considerations of the RTP specification, the RTP/SAVPF
profile, and the various RTP/RTCP extensions and RTP payload formats
that form the complete protocol suite described in this memo apply. We
do not believe there are any new security considerations resulting from
the combination of these various protocol extensions.The Extended Secure RTP Profile for Real-time
Transport Control Protocol (RTCP)-Based Feedback (RTP/SAVPF)
provides handling of fundamental issues by offering confidentiality,
integrity and partial source authentication. A mandatory to implement
media security solution is created by combing this secured RTP profile
and DTLS-SRTP keying as defined by Section 5.5 of.RTCP packets convey a Canonical Name (CNAME) identifier that is used
to associate RTP packet streams that need to be synchronised across
related RTP sessions. Inappropriate choice of CNAME values can be a
privacy concern, since long-term persistent CNAME identifiers can be
used to track users across multiple WebRTC calls. of this memo provides guidelines for generation of
untraceable CNAME values that alleviate this risk.The guidelines in apply when using variable
bit rate (VBR) audio codecs such as Opus (see for discussion of mandated audio codecs). The
guidelines in also apply, but are of lesser
importance, when using the client-to-mixer audio level header extensions
() or the mixer-to-client audio
level header extensions (). The use
of the encryption of the header extensions are RECOMMENDED, unless there
are known reasons, like RTP middleboxes or third party monitoring that
will greatly benefit from the information, and this has been expressed
using API or signalling. If further evidence are produced to show that
information leakage is significant from audio level indications, then
use of encryption needs to be mandated at that time.This memo makes no request of IANA.Note to RFC Editor: this section is to be removed on publication as
an RFC.The authors would like to thank Bernard Aboba, Harald Alvestrand,
Cary Bran, Charles Eckel, Christian Groves, Cullen Jennings, Dan
Romascanu, Martin Thomson, and the other members of the IETF RTCWEB
working group for their valuable feedback.