RTP TopologiesEricssonFarogatan 6SE-164 80 KistaSweden+46 10 714 82 87magnus.westerlund@ericsson.comVidyo433 Hackensack AveHackensackNJ07601USAstewe@stewe.orgThis document discusses point to point and multi-endpoint topologies
used in Real-time Transport Protocol (RTP)-based environments. In
particular, centralized topologies commonly employed in the video
conferencing industry are mapped to the RTP terminology.This document is updated with additional topologies and replaces RFC
5117.Real-time Transport Protocol (RTP)
topologies describe methods for interconnecting RTP entities and their
processing behavior of RTP and RTCP. This document tries to address past
and existing confusion, especially with respect to terms not defined in
RTP but in common use in the conversational communication industry, such
as the Multipoint Control Unit or MCU.When the Audio-Visual Profile with Feedback
(AVPF) was developed the main emphasis lay in the efficient
support of point to point and small multipoint scenarios without
centralized multipoint control. In practice, however, most multipoint
conferences operate utilizing centralized units referred to as MCUs.
MCUs may implement Mixer or Translator functionality (in RTP terminology), and signalling support. They
may also contain additional application layer functionality. This
document focuses on the media transport aspects of the MCU that can be
realized using RTP, as discussed below. Further considered are the
properties of Mixers and Translators, and how some types of deployed
MCUs deviate from these properties.This document also codifies new multipoint architectures that have
recently been introduced and which were not anticipated in RFC 5117,
thus this document replaces . These
architectures use scalable video coding and simulcasting, and their
associated centralized units are referred to as Selective Forwarding
Units (SFU). This codification provides a common information basis for
future discussion and specification work.The document's attempt to clarify and explain sections of the Real-time Transport Protocol (RTP) spec is
informal. It is not intended to update or change what is normatively
specified within RFC 3550.Any Source MulticastThe Extended RTP Profile for RTCP-based
FeedbackContributing SourceThe data transport to the next IP hopA device that is on the Path that media
travel between two endpointsMultipoint Control UnitThe concatenation of multiple links, resulting
in an end-to-end data transfer.Point to MultipointPoint to PointSelective Forwarding UnitSource-Specific MulticastSynchronization Source[Note to RFC editor: The following definitions have been taken from
draft-ietf-avtext-rtp-grouping-taxonomy-02 (taxonomy draft
henceforth). It is avtcore working group agreement to not delay the
publication of the topologies-update document through a dependency to
the taxonomy draft. If, however, the taxonomy draft and this draft are
in your work queue at the same time and there would be no significant
additional delay (through your schedule, normative reference
citations, or similar) in publishing both documents roughly in
parallel, it would be preferable to replace the definition language
with something like "as in [RFC YYYY]" where YYYY would be the RFC
number of the published taxonomy draft.]The following definitions have been taken from
draft-ietf-avtext-rtp-grouping-taxonomy-02, and are used in
capitalized form throughout the document.A Communication Session is an
association among group of participants communicating with each
other via a set of Multimedia Sessions.A single addressable entity sending or
receiving RTP packets. It may be decomposed into several
functional blocks, but as long as it behaves as a single RTP stack
entity it is classified as a single "Endpoint".A Media Source is the logical source
of a reference clock synchronized, time progressing, digital media
stream, called a Source Stream.A multimedia session is an
association among a group of participants engaged in the
communication via one or more RTP Sessions.This subsection defines several topologies that are relevant for
codec control but also RTP usage in other contexts. The section starts
with point to point cases, with or without middleboxes. Then follows a
number of different methods for establishing point to multipoint
communication. These are structured around the most fundamental enabler,
i.e., multicast, a mesh of connections, translators, mixers and finally
MCUs and SFUs. The section ends by discussing de-composited terminals,
asymmetric middlebox behaviors and combining topologies.The topologies may be referenced in other documents by a shortcut
name, indicated by the prefix "Topo-".For each of the RTP-defined topologies, we discuss how RTP, RTCP, and
the carried media are handled. With respect to RTCP, we also discuss the
handling of RTCP feedback messages as defined in and .Shortcut name: Topo-Point-to-PointThe Point to Point (PtP)
topology consists of two endpoints, communicating using
unicast. Both RTP and RTCP traffic are conveyed endpoint-to-endpoint,
using unicast traffic only (even if, in exotic cases, this unicast
traffic happens to be conveyed over an IP-multicast address).The main property of this topology is that A sends to B, and only
B, while B sends to A, and only A. This avoids all complexities of
handling multiple endpoints and combining the requirements stemming
from them. Note that an endpoint can still use multiple RTP
Synchronization Sources (SSRCs) in an RTP session. The number of RTP
sessions in use between A and B can also be of any number, subject
only to system level limitations like the number range of ports.RTCP feedback messages for the indicated SSRCs are communicated
directly between the endpoints. Therefore, this topology poses minimal
(if any) issues for any feedback messages. For RTP sessions which use
multiple SSRC per endpoint it can be relevant to implement support for
cross-reporting suppression as defined in "Sending
Multiple Media Streams in a Single RTP Session".This section discusses cases where two endpoints communicate but
have one or more middleboxes involved in the RTP session.Shortcut name: Topo-PtP-TranslatorTwo main categories of Translators can be distinguished;
Transport Translators and Media translators. Both Translator types
share common attributes that separate them from Mixers. For each RTP
stream that the Translator receives, it generates an individual RTP
stream in the other domain. A translator keeps the SSRC for an RTP
stream across the translation, whereas a Mixer can select a single
RTP stream from multiple received RTP streams (in cases like
audio/video switching), or send out an RTP stream composed of
multiple mixed media received in multiple RTP streams (in cases like
audio mixing or video tiling), but always under its own SSRC,
possibly using the CSRC field to indicate the source(s) of the
content. Mixers are more common in point to multipoint cases than in
PtP. The reason is that in PtP use cases the primary focus of a
middlebox is enabling interoperability, between otherwise
non-interoperable endpoints, such as transcoding to a codec the
receiver supports, which can be done by a media translator.As specified in Section 7.1 of , the SSRC
space is common for all participants in the RTP session, independent
of on which side of the Translator the session resides. Therefore,
it is the responsibility of the endpoints (as the RTP session
participants) to run SSRC collision detection, and the SSRC is thus
a field the Translator cannot change. Any SDES information
associated with a SSRC or CSRC also needs to be forwarded between
the domains for any SSRC/CSRC used in the different domains.A Translator commonly does not use an SSRC of its own, and is not
visible as an active participant in the RTP session. One reason to
have its own SSRC is when a Translator acts as a quality monitor
that sends RTCP reports and therefore is required to have an SSRC.
Another example is the case when a Translator is prepared to use
RTCP feedback messages. This may, for example, occur in a translator
configured to detect packet loss of important video packets and
wants to trigger repair by the media sending endpoint, by sending
feedback messages. While such feedback could use the SSRC of the
target for the translator (the receiving endpoint), this in turn
would require translation of the targets RTCP reports to make them
consistent. It may be simpler to expose an additional SSRC in the
session. The only concern is endpoints failing to support the full
RTP specification may have issues with multiple SSRCs reporting on
the RTP streams sent by that endpoint, as this use case may be
viewed as excotic by implementers.In general, a Translator implementation should consider which
RTCP feedback messages or codec-control messages it needs to
understand in relation to the functionality of the Translator
itself. This is completely in line with the requirement to also
translate RTCP messages between the domains.There exist a number of different types of middleboxes that
might be inserted between two endpoints on the transport level,
e.g., to perform changes on the IP/UDP headers, and are,
therefore, basic transport translators. These middleboxes come in
many variations including NAT
traversal by pinning the media path to a public address domain
relay, network topologies where the RTP stream is required to pass
a particular point for audit by employing relaying, or preserving
privacy by hiding each peer's transport addresses to the other
party. Other protocols or functionalities that provide this
behavior are TURN servers, Session
Border Gateways and Media Processing Nodes with media anchoring
functionalities.A common element in these functions is that they are normally
transparent at the RTP level, i.e., they perform no changes on any
RTP or RTCP packet fields and only affect the lower layers. They
may affect, however, the path the RTP and RTCP packets are routed
between the endpoints in the RTP session, and thereby indirectly
affect the RTP session. For this reason, one could believe that
transport translator-type middleboxes do not need to be included
in this document. This topology, however, can raise additional
requirements in the RTP implementation and its interactions with
the signalling solution. Both in signalling and in certain RTCP
fields, network addresses other than those of the relay can occur
since B has a different network address than the relay (T).
Implementations that cannot support this will also not work
correctly when endpoints are subject to NAT.The transport relay implementations also have to take into
account security considerations. In particular, source address
filtering of incoming packets is usually important in relays, to
prevent attackers to inject traffic into a session, which one peer
may, in the absence fo adequate security in the relay, think it
comes from the other peer.Transport Translators (Topo-Trn-Translator) do not modify the
RTP stream itself, but are concerned with transport parameters.
Transport parameters, in the sense of this section, comprise the
transport addresses (to bridge different domains such unicast to
multicast) and the media packetization to allow other transport
protocols to be interconnected to a session (in gateways).Translators that bridge between different protocol worlds need
to be concerned about the mapping of the SSRC/CSRC (Contributing
Source) concept to the non-RTP protocol. When designing a
Translator to a non-RTP-based media transport, an important
consideration is how to handle different sources and their
identities. This problem space is not discussed henceforth.Of the transport Translators, this memo is primarily interested
in those that use RTP on both sides, and this is assumed
henceforth.The most basic transport translators that operate below the RTP
level were already discussed in .Media Translators (Topo-Media-Translator) modify the media
inside the RTP stream. This process is commonly known as
transcoding. The modification of the media can be as small as
removing parts of the stream, and it can go all the way to a full
decoding and re-encoding (down to the sample level or equivalent)
utilizing a different media codec. Media Translators are commonly
used to connect endpoints without a common interoperability point
in the media encoding.Stand-alone Media Translators are rare. Most commonly, a
combination of Transport and Media Translator is used to translate
both the media and the transport aspects of the RTP stream
carrying the media between two transport domains.When media translation occurs, the Translator's task regarding
handling of RTCP traffic becomes substantially more complex. In
this case, the Translator needs to rewrite endpoint B's RTCP
Receiver Report before forwarding them to endpoint A. The
rewriting is needed as the RTP stream received by B is not the
same RTP stream as the other participants receive. For example,
the number of packets transmitted to B may be lower than what A
sends, due to the different media format and data rate. Therefore,
if the Receiver Reports were forwarded without changes, the
extended highest sequence number would indicate that B were
substantially behind in reception, while most likely it would not
be. Therefore, the Translator must translate that number to a
corresponding sequence number for the stream the Translator
received. Similar requirements exists for most other fields in the
RTCP Receiver Reports.A media Translator may in some cases act on behalf of the
"real" source (the endpoint originally sending the media to the
Translator) and respond to RTCP feedback messages. This may occur,
for example, when a receiving endpoint requests a bandwidth
reduction, and the media Translator has not detected any
congestion or other reasons for bandwidth reduction between the
sending endpoint and itself. In that case, it is sensible that the
media Translator reacts to codec control messages itself, for
example, by transcoding to a lower media rate.A variant of translator behaviour worth pointing out is the one
depicted in of an
endpoint A sending a RTP stream containing media (only) to B. On
the path there is a device T that on A's behalf manipulates the
RTP streams. One common example is that T adds a second RTP stream
containing Forward Error Correction (FEC) information in order to
protect A's (non FEC-protected) RTP stream. In this case, T needs
to semantically bind the new FEC RTP stream to A's media-carrying
RTP stream, for example by using the same CNAME as A.there may also be cases where information is added into the
original RTP stream, while leaving most or all of the original RTP
packets intact (with the exception of certain RTP header fields,
such as the sequence number). One example is the injection of
meta-data into the RTP stream, carried in their own RTP
packets.Similarly, a Media Translator can sometimes remove information
from the RTP stream, while otherwise leaving the remaining RTP
packets unchanged (again with the exception of certain RTP header
fields).Either type of functionality where T manipulates the RTP
stream, or adds an accompanying RTP stream, on behalf of A is also
covered under the media translator definition.There exist middleboxes that interconnect two endpoints A and B
through themselves (MB), but not by being part of a common RTP
session. They establish instead two different RTP sessions, one
between A and the middlebox and another between the middlebox and B.
This topology is called Topo-Back-To-BackThe middlebox acts as an application-level gateway and bridges
the two RTP sessions. This bridging can be as basic as forwarding
the RTP payloads between the sessions, or more complex including
media transcoding. The difference of this topology relative to the
single RTP session context is the handling of the SSRCs and the
other session-related identifiers, such as CNAMEs. With two
different RTP sessions these can be freely changed and it becomes
the middlebox's respnsibility to maintain the correct relations.The signalling or other above-RTP level functionalities
referencing RTP streams may be what is most impacted by using two
RTP sessions and changing identifiers. The structure with two RTP
sessions also puts a congestion control requirement on the
middlebox, because it becomes fully responsible for the media stream
it sources into each of the sessions.Adherence to congestion control can be solved locally on each of
the two segments, or by bridging statistics from the receiving
endpoint through the middlebox to the sending endpoint. From an
implementation point, however, the latter requires dealing with a
number of inconsistencies. First, packet loss must be detected for
an RTP stream sent from A to the middlebox, and that loss must be
reported through a skipped sequence number in the RTP stream from
the middlebox to B. This coupling and the resulting inconsistencies
are conceptually easier to handle when considering the two RTP
streams as belonging to a single RTP session.Multicast is an IP layer functionality that is available in some
networks. Two main flavors can be distinguished: Any Source Multicast (ASM) where any multicast
group participant can send to the group address and expect the packet
to reach all group participants; and Source
Specific Multicast (SSM), where only a particular IP host sends
to the multicast group. Both these models are discussed below in their
respective sections.Shortcut name: Topo-ASM (was Topo-Multicast)Point to Multipoint (PtM) is defined here as using a multicast
topology as a transmission model, in which traffic from any
multicast group participant reaches all the other multicast group
participants, except for cases such as:packet loss, orwhen a multicast group participant does not wish to receive
the traffic for a specific multicast group and, therefore, has
not subscribed to the IP multicast group in question. This
scenario can occur, for example, where a multimedia session is
distributed using two or more multicast groups and a multicast
group participant is subscribed only to a subset of these
sessions.In the above context, "traffic" encompasses both RTP and RTCP
traffic. The number of multicast group participants can vary between
one and many, as RTP and RTCP scale to very large multicast groups
(the theoretical limit of the number of participants in a single RTP
session is in the range of billions). The above can be realized
using Any Source Multicast (ASM).For feedback usage, it is useful to define a "small multicast
group" as a group where the number of multicast group participants
is so low (and other factors such as the connectivity is so good)
that it allows the participants to use early or immediate feedback,
as defined in AVPF. Even when the
environment would allow for the use of a small multicast group, some
applications may still want to use the more limited options for RTCP
feedback available to large multicast groups, for example when there
is a likelihood that the threshold of the small multicast group (in
terms of multicast group participants) may be exceeded during the
lifetime of a session.RTCP feedback messages in multicast reach, like media data, every
subscriber (subject to packet losses and multicast group
subscription). Therefore, the feedback suppression mechanism
discussed in is typically required. Each
individual endpoint that is a multicast group participant needs to
process every feedback message it receives, not only to determine if
it is affected or if the feedback message applies only to some other
endpoint, but also to derive timing restrictions for the sending of
its own feedback messages, if any.In Any Source Multicast, any of the multicast group participants
can send to all the other multicast group participants, by sending a
packet to the multicast group. In contrast, Source Specific Multicast refers to scenarios where only a single source
(Distribution Source) can send to the multicast group, creating a
topology that looks like the one below:In the SSM topology a
number of RTP sending endpoints (RTP sources henceforth) (1 to M)
are allowed to send media to the SSM group. These sources send media
to a dedicated distribution source, which forwards the RTP streams
to the multicast group on behalf of the original RTP sources. The
RTP streams reach the receiving endpoints (Receivers henceforth)
(R(1) to R(n)). The Receivers' RTCP messages cannot be sent to the
multicast group, as the SSM multicast group by definition has only a
single IP sender. To support RTCP, an RTP
extension for SSM was defined. It uses unicast transmission
to send RTCP from each of the receivers to one or more Feedback
Targets (FT). The feedback targets relay the RTCP unmodified, or
provide a summary of the participants RTCP reports towards the whole
group by forwarding the RTCP traffic to the distribution source.
only shows a single feedback
target integrated in the distribution source, but for scalability
the FT can be distributed and each instance can have responsibility
for sub-groups of the receivers. For summary reports, however, there
typically must be a single feedback target aggregating all the
summaries to a common message to the whole receiver group.The RTP extension for SSM specifies how feedback (both reception
information and specific feedback events) are handled. The more
general problems associated with the use of multicast, where
everyone receives what the distribution source sends needs to be
accounted for.Aforementioned situation results in common behavior for RTP
multicast:Multicast applications often use a group of RTP sessions, not
one. Each endpoint needs to be a member of most or all of these
RTP sessions in order to perform well.Within each RTP session, the number of media sinks is likely
to be much larger than the number of RTP sources.Multicast applications need signalling functions to identify
the relationships between RTP sessions.Multicast applications need signalling functions to identify
the relationships between SSRCs in different RTP sessions.All multicast configurations share a signalling requirement: all
of the endpoints need to have the same RTP and payload type
configuration. Otherwise, endpoint A could, for example, be using
payload type 97 to identify the video codec H.264, while endpoint B
would identify it as MPEG-2, with unpredicatble but almost certainly
not visually pleasing results.Security solutions for this type of group communications are also
challenging. First, the key-management and the security protocol
must support group communication. Source authentication becomes more
difficult and requires specialized solutions. For more discussion on
this please review Options for Securing RTP
Sessions."Unicast-Based Rapid Acquisition of
Multicast RTP Sessions" results in additional extensions to
SSM Topology.The Rapid acquisition extension allows an endpoint joining an SSM
multicast session to request media starting with the last sync-point
(from where media can be decoded without requiring context
established by the decoding of prior packets) to be sent at high
speed until such time where, after decoding of these burst-delivered
media packets, the correct media timing is established, i.e. media
packets are received within adequate buffer intervals for this
application. This is accomplished by first establishing a unicast
PtP RTP session between the Burst/Retransmission Source (BRS, ) and the RTP Receiver. The unicast session is
used to transmit cached packets from the multicast group at higher
then normal speed in order to synchronize the receiver to the
ongoing multicast RTP stream. Once the RTP receiver and its decoder
have caught up with the multicast session's current delivery, the
receiver switches over to receiving directly from the multicast
group. In many deployed application, the (still existing) PtP RTP
session is used as a repair channel, i.e., for RTP Retransmission
traffic of those packets that were not received from the multicast
group.Shortcut name: Topo-MeshBased on the RTP session definition, it is clearly possible to have
a joint RTP session involving three or more endpoints over multiple
unicast transport flows, like the joint three endpoint session
depicted above. In this case, A needs to send its RTP streams and RTCP
packets to both B and C over their respective transport flows. As long
as all endpoints do the same, everyone will have a joint view of the
RTP session.This topology does not create any additional requirements beyond
the need to have multiple transport flows associated with a single RTP
session. Note that an endpoint may use a single local port to receive
all these transport flows (in which case the sending port, IP address,
or SSRC can be used to demultiplex), or it might have separate local
reception ports for each of the endpoints.A joint RTP session from endpoint A's perspective for the Mesh
depicted in with a joint RTP session have
multiple transport flows, here enumerated as UDP1 and UDP2. However,
there is only one RTP session (RTP1). The Media Source (CAM) is
encoded and transmitted over the SSRC (AV1) across both transport
layers. However, as this is a joint RTP session, the two streams must
be the same. Thus, an congestion control adaptation needed for the
paths A to B and A to C needs to use the most restricting path's
properties.An alternative structure for establishing the above topology is to
use independent RTP sessions between each pair of peers, i.e., three
different RTP sessions. In some scenarios, the same RTP stream may be
sent from the transmitting endpoint, however it also supports local
adaptation taking place in one or more of the RTP streams, rendering
them non-identical.Lets review the topology when independent RTP sessions are used,
from A's perspective in by
considering both how the media is handled and the RTP sessions that
are set-up in . A's microphone
is captured and the audio is fed into two different encoder instances,
each being associated with two independent RTP sessions (RTP1 and
RTP2). The SSRCs (AA1 and AA2) in each RTP session are completely
independent and the media bit-rate produced by the encoders can also
be tuned differently to address any congestion control requirements
differing for the paths A to B compared to A to C.From a topologies viewpoint, an important difference exists in the
behavior around RTCP. First, when a single RTP session spans all three
endpoints A, B, and C, and their connecting RTP streams, a common RTCP
bandwidth is calculated and used for this single joint session. In
contrast, when there are multiple independent RTP sessions, each RTP
session has its local RTCP bandwidth allocation.Further, when multiple sessions are used, endpoints not directly
involved in a session do not have any awareness of the conditions in
those sessions. For example, in the case of the three endpoint
configuration in , endpoint A has no
awareness of the conditions occurring in the session between endpoints
B and C (whereas, if a single RTP session were used, it would have
such awareness).Loop detection is also affected. With independent RTP sessions, the
SSRC/CSRC cannot be used to determine when an endpoint receives its
own media stream, or a mixed media stream including its own media
stream (a condition known as a loop). The identification of loops and,
in most cases, their avoidance, has to be achieved by other means, for
example through signaling or the use of an RTP external name space
binding SSRC/CSRC among any communicating RTP sessions in the
mesh.This section discusses some additional usages related to point to
multipoint of Translators compared to the point to point only cases in
.Shortcut name: Topo-PtM-Trn-TranslatorThis section discusses Transport Translator only usages to enable
multipoint sessions. depicts an example
of a Transport Translator performing at least IP address
translation. It allows the (non-multicast-capable) endpoints B and D
to take part in an any source multicast session involving endpoints
A and C, by having the Translator forward their unicast traffic to
the multicast addresses in use, and vice versa. It must also forward
B's traffic to D, and vice versa, to provide each of B and D with a
complete view of the session.Another Translator scenario is depicted in . The Translator in this case
connects multiple endpoints through unicast. This can be implemented
using a very simple transport Translator which, in this document, is
called a relay. The relay forwards all traffic it receives, both RTP
and RTCP, to all other endpoints. In doing so, a multicast network
is emulated without relying on a multicast-capable network
infrastructure.For RTCP feedback this results in a similar set of considerations
to those described in the ASM RTP topology. It also puts some
additional signalling requirements onto the session establishment;
for example, a common configuration of RTP payload types is
required.Transport translators and relays should always consider
implementing source address filtering, to prevent attackers to
inject traffic using the listening ports on the translator. The
translator can, however, go one step further, and especially if
explicit SSRC signalling is used, prevent endpoints to send SSRCs
other than its own (that are, for example, used by other
participants in the session). This can improve the security
properties of the session, despite the use of group keys that on
cryptographic level allows anyone to impersonate another in the same
RTP session.A Translator that doesn't change the RTP/RTCP packets content can
be operated without the requiring it to have access to the security
contexts used to protect the RTP/RTCP traffic between the
participants.In the context of multipoint communications a Media Translator is
not providing new mechanisms to establish a multipoint session. It
is more of an enabler, or facilitator, that ensures a given endpoint
or a defined sub-set of endpoints can participate in the
session.If endpoint B in
were behind a limited network path, the Translator may perform media
transcoding to allow the traffic received from the other endpoints
to reach B without overloading the path. This transcoding can help
the other endpoints in the multicast part of the session, by not
requiring the quality transmitted by A to be lowered to the bitrates
that B is actually capable of receiving (and vice versa).Shortcut name: Topo-MixerA Mixer is a middlebox that aggregates multiple RTP streams that
are part of a session by generating one or more new RTP streams and,
in most cases, by manipulating the media data. One common application
for a Mixer is to allow a participant to receive a session with a
reduced amount of resources.A Mixer can be viewed as a device terminating the RTP streams
received from other endpoints in the same RTP session. Using the media
data carried in the received RTP streams, a Mixer generates derived
RTP streams that are sent to the receiving endpoints.The content that the Mixer provides is the mixed aggregate of what
the Mixer receives over the PtP or PtM paths, which are part of the
same Communication Session.The Mixer creates the Media Source and the source RTP stream just
like an endpoint, as it mixes the content (often in the uncompressed
domain) and then encodes and packetizes it for transmission to a
receiving endpoint. The CSRC Count (CC) and CSRC fields in the RTP
header can be used to indicate the contributors to the newly generated
RTP stream. The SSRCs of the to-be-mixed streams on the Mixer input
appear as the CSRCs at the Mixer output. That output stream uses a
unique SSRC that identifies the Mixer's stream. The CSRC should be
forwarded between the different endpoints to allow for loop detection
and identification of sources that are part of the Communication
Session. Note that Section 7.1 of RFC 3550 requires the SSRC space to
be shared between domains for these reasons. This also implies that
any SDES information normally needs to be forwarded across the
mixer.The Mixer is responsible for generating RTCP packets in accordance
with its role. It is an RTP receiver and should therefore send RTCP
receiver reports for the RTP streams it receives and terminates. In
its role as an RTP sender, it should also generate RTCP sender reports
for those RTP streams it sends. As specified in Section 7.3 of RFC
3550, a Mixer must not forward RTCP unaltered between the two
domains.The Mixer depicted in is involved in
three domains that need to be separated: the any source multicast
network (including endpoints A and C), endpoint B, and endpoint D.
Assuming all four endpoints in the conference are interested in
receiving content from all other endpoints, the Mixer produces
different mixed RTP streams for B and D, as the one to B may contain
content received from D, and vice versa. However, the Mixer may only
need one SSRC per media type in each domain where it is the receiving
entity and transmitter of mixed content.In the multicast domain, a Mixer still needs to provide a mixed
view of the other domains. This makes the Mixer simpler to implement
and avoids any issues with advanced RTCP handling or loop detection,
which would be problematic if the Mixer were providing non-symmetric
behavior. Please see for more
discussion on this topic. The mixing operation, however, in each
domain could potentially be different.A Mixer is responsible for receiving RTCP feedback messages and
handling them appropriately. The definition of "appropriate" depends
on the message itself and the context. In some cases, the reception of
a codec-control message by the Mixer may result in the generation and
transmission of RTCP feedback messages by the Mixer to the endpoints
in the other domain(s). In other cases, a message is handled by the
Mixer locally and therefore not forwarded to any other domain.When replacing the multicast network in (to the left of the Mixer) with individual
unicast paths as depicted in , the
Mixer model is very similar to the one discussed in below. Please see the discussion in about the differences between these two
models.We now discuss in more detail the different mixing operations that
a mixer can perform and how they can affect RTP and RTCP behavior.The media mixing mixer is likely the one that most think of when
they hear the term "mixer". Its basic mode of operation is that it
receives RTP streams from several endpoints and selects the
stream(s) to be included in a media-domain mix. The selection can be
through static configuration or by dynamic, content dependent means
such as voice activation. The mixer then creates a single outgoing
RTP stream from this mix.The most commonly deployed media mixer is probably the audio
mixer, used in voice conferencing, where the output consists of a
mixture of all the input audio signals; this needs minimal
signalling to be successfully set up. From a signal processing
viewpoint, audio mixing is relatively straightforward and commonly
possible for a reasonable number of endpoints. Assume, for example,
that one wants to mix N streams from N different endpoints. The
mixer needs to decode those N streams, typically into the sample
domain, and then produce N or N+1 mixes. Different mixes are needed
so that each contributing source gets a mix of all other sources
except its own, as this would result in an echo. When N is lower
than the number of all endpoints, one may produce a mix of all N
streams for the group that are currently not included in the mix,
thus N+1 mixes. These audio streams are then encoded again, RTP
packetized and sent out. In many cases, audio level normalization,
noise suppression, and similar signal processing steps are also
required or desirable before the actual mixing process
commences.In video, the term "mixing" has a different interpretation than
audio. It is commonly used to refer to the process of spatially
combining contributed video streams, which is also known as
"tiling". The reconstructed, appropriately scaled down videos can be
spatially arranged in a set of tiles, each tile containing the video
from an endpoint (typically showing a human participant). Tiles can
be of different sizes, so that, for example, a particularly
important participant, or the loudest speaker, is being shown on in
larger tile than other participants. A self-view picture can be
included in the tiling, which can either be locally produced or be a
feedback from a mixer-received and reconstructed video image. Such
remote loopback allows for confidence monitoring, i.e., it enables
the participant to see himself/herself in the same quality as other
participants see him/her. The tiling normally operates on
reconstructed video in the sample domain. The tiled image is
encoded, packetized, and sent by the mixer to the receiving
endpoints. It is possible that a middlebox with media mixing duties
contains only a single mixer of the aforementioned type, in which
case all participants necessarily see the same tiled video, even if
it is being sent over different RTP streams. More common, however,
are mixing arrangement where an individual mixer is available for
each outgoing port of the middlebox, allowing individual
compositions for each receiving endpoint (a feature commonly
referred to as personalized layout).One problem with media mixing is that it consumes both large
amounts of media processing resources (for the decoding and mixing
process in the uncompressed domain) and encoding resources (for the
encoding of the mixed signal). Another problem is the quality
degradation created by decoding and re-encoding the media, which is
the result of the lossy nature of most commonly used media codecs. A
third problem is the latency introduced by the media mixing, which
can be substantial and annoyingly noticeable in case of video, or in
case of audio if that mixed audio is lip-sychronized with high
latency video. The advantage of media mixing is that it is
straightforward for the endpoints to handle the single media stream
(which includes the mixed aggregate of many sources), as they don't
need to handle multiple decodings, local mixing and composition. In
fact, mixers were introduced in pre-RTP times so that legacy, single
stream receiving endpoints (that, in some protocol environments,
actually didn't need to be aware of the multipoint nature of the
conference) could successfully participate in what a user would
recognize as a multiparty video conference.From an RTP perspective media mixing can be a very simple
process, as can be seen in . The
mixer presents one SSRC towards the receiving endpoint, e.g., MA1 to
Peer A, where the associated stream is the media mix of the other
endpoints. As each peer, in this example, receives a different
version of a mix from the mixer, there is no actual relation between
the different RTP sessions in terms of actual media or transport
level information. There are, however, common relationships between
RTP1-RTP3, namely SSRC space and identity information. When A
receives the MA1 stream which is a combination of BA1 and CA1
streams, the mixer may include CSRC information in the MA1 stream to
identify the contributing source BA1 and CA1, allowing the receiver
to identify the contributing sources even if this were not possible
through the media itself or through other signaling means.The CSRC has, in turn, utility in RTP extensions, like the Mixer to Client audio levels RTP header
extension. If the SSRCs from the endpoint to mixer paths are
used as CSRCs in another RTP session, then RTP1, RTP2 and RTP3
become one joint session as they have a common SSRC space. At this
stage, the mixer also needs to consider which RTCP information it
needs to expose in the different paths. In the above scenario, a
mixer would normally expose nothing more than the Source Description
(SDES) information and RTCP BYE for a CSRC leaving the session. The
main goal would be to enable the correct binding against the
application logic and other information sources. This also enables
loop detection in the RTP session.Media switching mixers are used in limited functionality
scenarios where no, or only very limited, concurrent presentation of
multiple sources is required by the application, to more complex
multi-stream usages with receiver mixing or tiling, including
combined with simulcast and/or scalability between source and mixer.
An RTP Mixer based on media switching avoids the media decoding and
encoding operations in the mixer, as it conceptually forwards the
encoded media stream as it was being sent to the mixer. It does not
avoid, however, the decryption and re-encryption cycle as it
rewrites RTP headers. Forwarding media (in contrast to
reconstructing-mixing-encoding media) reduces the amount of
computational resources needed in the mixer and increases the media
quality (both in terms of fidelity and reduced latency).A media switching mixer maintains a pool of SSRCs representing
conceptual or functional RTP streams that the mixer can produce.
These RTP streams are created by selecting media from one of the RTP
streams received by the mixer and forwarded to the peer using the
mixer's own SSRCs. The mixer can switch between available sources if
that is required by the concept for the source, like the currently
active speaker. Note that the mixer, in most cases, still needs to
perform a certain amount of media processing, as many media formats
do not allow to "tune into" the stream at arbitrary points in their
bitstream.To achieve a coherent RTP stream from the mixer's SSRC, the mixer
needs to rewrite the incoming RTP packet's header. First the SSRC
field must be set to the value of the Mixer's SSRC. Second, the
sequence number must be the next in the sequence of outgoing packets
it sent. Third, the RTP timestamp value needs to be adjusted using
an offset that changes each time one switches media source. Finally,
depending on the negotiation of the RTP payload type, the value
representing this particular RTP payload configuration may have to
be changed if the different endpoint-to-mixer paths have not arrived
on the same numbering for a given configuration. This also requires
that the different endpoints support a common set of codecs,
otherwise media transcoding for codec compatibility would still be
required.We now consider the operation of a media switching mixer that
supports a video conference with six participating endpoints (A-F)
where the two most recent speakers in the conference are shown to
each receiving endpoint. The mixer has thus two SSRCs sending video
to each peer, and each peer is capable of locally handling two video
streams simultaneously.The Media Switching RTP mixer can, similarly to the Media Mixing
Mixer, reduce the bit-rate required for media transmission towards
the different peers by selecting and forwarding only a sub-set of
RTP streams it receives from the sending endpoints. In cases the
mixer receives simulcast transmissions or a scalable encoding of the
media source, the mixer has more degrees of freedom to select
streams or sub-sets of stream to forward to a receiving endpoint,
both based on transport or endpoint restrictions as well as
application logic.To ensure that a media receiver in an endpoint can correctly
decode the media in the RTP stream after a switch, a codec that uses
temporal prediction needs to start its decoding from independent
refresh points, or points in the bitstream offering similar
functionality (like "dirty refresh points"). For some codecs, for
example frame based speech and audio codecs, this is easily achieved
by starting the decoding at RTP packet boundaries, as each packet
boundary provides a refresh point (assuming proper packetization on
the encoder side). For other codecs, particularly in video, refresh
points are less common in the bitstream or may not be present at all
without an explicit request to the respective encoder. The Full Intra Request RTCP codec control
message has been defined for this purpose.In this type of mixer one could consider to fully terminate the
RTP sessions between the different endpoint and mixer paths. The
same arguments and considerations as discussed in need to be taken into consideration and apply
here.Another method for handling media in the RTP mixer is to "project",
or make available, all potential RTP sources (SSRCs) into a
per-endpoint, independent RTP session. The middlebox can select which
of the potential sources that are currently actively transmitting
media will be sent to each of the endpoints. This is similar to the
media switching Mixer but has some important differences in RTP
details.In the six endpoint conference depicted above in one can see that endpoint A is aware
of five incoming SSRCs, BV1-FV1. If this middlebox intends to have a
similar behavior as in where the
mixer provides the endpoints with the two latest speaking endpoints,
then only two out of these five SSRCs need concurrently transmit media
to A. As the middlebox selects the source in the different RTP
sessions that transmit media to the endpoints, each RTP stream
requires rewriting of certain RTP header fields when being projected
from one session into another. In particular, the sequence number
needs to be consecutively incremented based on the packet actually
being transmitted in each RTP session. Therefore, the RTP sequence
number offset will change each time a source is turned on in a RTP
session. The timestamp (possibly offset) stays the same.As the RTP sessions are independent, the SSRC numbers used can also
be handled independently, thereby bypassing the requirement for SSRC
collision detection and avoidance. On the other hand, tools such as
remapping tables between the RTP sessions are required. For example,
the RTP stream that is being sent by endpoint B to the middlebox (BV1)
may use an SSRC value of 12345678. When that RTP stream is sent to
endpoint F by the middlebox, it can use any SSRC value, e.g. 87654321.
As a result, each endpoint may have a different view of the
application usage of a particular SSRC. Any RTP level identity
information, such as SDES items also needs to update the SSRC
referenced, if the included SDES items are intended to be global. Thus
the application must not use SSRC as references to RTP streams when
communicating with other peers directly. This also affects loop
detection which will fail to work, as there is no common namespace and
identities across the different legs in the communication session on
RTP level. Instead this responsibility falls onto higher layers.The middlebox is also responsible to receive any RTCP codec control
requests coming from an endpoint, and decide if it can act on the
request locally or needs to translate the request into the RTP session
that contains the media source. Both endpoints and the middlebox need
to implement conference related codec control functionalities to
provide a good experience. Commonly used are Full Intra Request to
request from the media source to provide switching points between the
sources, and Temporary Maximum Media Bit-rate Request (TMMBR) to
enable the middlebox to aggregate congestion control responses towards
the media source so to enable it to adjust its bit-rate (obviously
only in case the limitation is not in the source to middlebox
link).The selective forwarding middlebox has been introduced in recently
developed videoconferencing systems in conjunction with, and to
capitalize on, scalable video coding as well as simulcasting. An
example of scalable video coding is Annex G of H.264, but other
codecs, including H.264 AVC and VP8 also exhibit scalability, albeit
only in the temporal dimension. In both scalable coding and simulcast
cases the video signal is represented by a set of two or more
bitstreams, providing a corresponding number of distinct fidelity
points. The middlebox selects which parts of a scalable bitstream (or
which bitstream, in the case of simulcasting) to forward to each of
the receiving endpoints. The decision may be driven by a number of
factors, such as available bit rate, desired layout, etc. Contrary to
transcoding MCUs, these "Selective Forwarding Units" (SFUs) have
extremely low delay, and provide features that are typically
associated with high-end systems (personalized layout, error
localization) without any signal processing at the middlebox. They are
also capable of scaling to a large number of concurrent users,
and--due to their very low delay--can also be cascaded.This version of the middlebox also puts different requirements on
the endpoint when it comes to decoder instances and handling of the
RTP streams providing media. As each projected SSRC can, at any time,
provide media, the endpoint either needs to be able to handle as many
decoder instances as the middlebox received, or have efficient
switching of decoder contexts in a more limited set of actual decoder
instances to cope with the switches. The application also gets more
responsibility to update how the media provided is to be presented to
the user.Note that this topology could potentially be seen as a media
translator which include an on/off logic as part of its media
translation. The main difference would be a common global SSRC space
in the case of the Media Translator and the mapped one used in the
above. It also has mixer aspects, as the streams it provides are not
basically translated version, but instead they have conceptual
property assigned to them. Thus this topology appears to be some
hybrid between the translator and mixer model.The differences between selective forwarding middlebox and a switching mixer are minor, and
they share most properties. The above requirement on having a large
number of decoding instances or requiring efficient switching of
decoder contexts, are one point of difference. The other is how the
identification is performed, where the Mixer uses CSRC to provide
information on what is included in a particular RTP stream that
represent a particular concept. Selective forwarding gets the source
information through the SSRC, and instead have to use other mechanism
to make clear the streams current purpose.Shortcut name: Topo-Video-switch-MCUThis PtM topology was popular in early implementations of
multipoint videoconferencing systems due to its simplicity, and the
corresponding middlebox design has been known as a "video switching
MCU". The more complex RTCP-terminating MCUs, discussed in the next
section, became the norm, however, when technology allowed
implementations at acceptable costs.A video switching MCU forwards to a participant a single media
stream, selected from the available streams. The criteria for
selection are often based on voice activity in the audio-visual
conference, but other conference management mechanisms (like
presentation mode or explicit floor control) are known to exist as
well.The video switching MCU may also perform media translation to
modify the content in bit-rate, encoding, or resolution. However, it
still may indicate the original sender of the content through the
SSRC. In this case, the values of the CC and CSRC fields are
retained.If not terminating RTP, the RTCP Sender Reports are forwarded for
the currently selected sender. All RTCP Receiver Reports are freely
forwarded between the endpoints. In addition, the MCU may also
originate RTCP control traffic in order to control the session and/or
report on status from its viewpoint.The video switching MCU has most of the attributes of a Translator.
However, its stream selection is a mixing behavior. This behavior has
some RTP and RTCP issues associated with it. The suppression of all
but one RTP stream results in most participants seeing only a subset
of the sent RTP streams at any given time, often a single RTP stream
per conference. Therefore, RTCP Receiver Reports only report on these
RTP streams. Consequently, the endpoints emitting RTP streams that are
not currently forwarded receive a view of the session that indicates
their RTP streams disappear somewhere en route. This makes the use of
RTCP for congestion control, or any type of quality reporting, very
problematic.To avoid the aforementioned issues, the MCU needs to implement two
features. First, it needs to act as a Mixer (see ) and forward the selected RTP stream under
its own SSRC and with the appropriate CSRC values. Second, the MCU
needs to modify the RTCP RRs it forwards between the domains. As a
result, it is recommended that one implement a centralized video
switching conference using a Mixer according to RFC 3550, instead of
the shortcut implementation described here.Shortcut name: Topo-RTCP-terminating-MCUIn this PtM scenario, each endpoint runs an RTP point-to-point
session between itself and the MCU. This is a very commonly deployed
topology in multipoint video conferencing. The content that the MCU
provides to each participant is either:a selection of the content received from the other endpoints,
orthe mixed aggregate of what the MCU receives from the other PtP
paths, which are part of the same Communication Session.In case (a), the MCU may modify the content in terms of bit-rate,
encoding format, or resolution. No explicit RTP mechanism is used to
establish the relationship between the original RTP stream of the
media being sent RTP stream the MCU sends. In other words, the
outgoing RTP streams typically use a different SSRC, and may well use
a different payload type (PT), even if this different PT happens to be
mapped to the same media type. This is a result of the individually
negotiated RTP session for each endpoint.In case (b), the MCU is the Media Source and generates the Source
RTP Stream as it mixes the received content and then encodes and
packetizes it for transmission to an endpoint. According to RTP, the SSRC of the contributors are to be
signalled using the CSRC/CC mechanism. In practice, today, most
deployed MCUs do not implement this feature. Instead, the
identification of the endpoints whose content is included in the
Mixer's output is not indicated through any explicit RTP mechanism.
That is, most deployed MCUs set the CSRC Count (CC) field in the RTP
header to zero, thereby indicating no available CSRC information, even
if they could identify the original sending endpoints as suggested in
RTP.The main feature that sets this topology apart from what RFC 3550
describes is the breaking of the common RTP session across the
centralized device, such as the MCU. This results in the loss of
explicit RTP-level indication of all participants. If one were using
the mechanisms available in RTP and RTCP to signal this explicitly,
the topology would follow the approach of an RTP Mixer. The lack of
explicit indication has at least the following potential
problems:Loop detection cannot be performed on the RTP level. When
carelessly connecting two misconfigured MCUs, a loop could be
generated.There is no information about active media senders available in
the RTP packet. As this information is missing, receivers cannot
use it. It also deprives the client of information related to
currently active senders in a machine-usable way, thus preventing
clients from indicating currently active speakers in user
interfaces, etc.Note that many/most deployed MCUs (and video conferencing
endpoints) rely on signalling layer mechanisms for the identification
of the contributing sources, for example, a SIP
conferencing package. This alleviates, to some extent, the
aforementioned issues resulting from ignoring RTP's CSRC
mechanism.Shortcut name: Topo-Split-TerminalIn some applications, for example in some telepresence systems,
terminals may be not integrated into a single functional unit, but
composed of more than one subunits. For example, a telepresence room
terminal employing multiple cameras and monitors may consist of
multiple video conferencing subunits, each capable of handling a
single camera and monitor. Another example would be a video
conferencing terminal in which audio is handled by one subunit, and
video by another. Each of these subunits uses its own physical network
interface (for example: Ethernet jack) and network address.The various (media processing) subunits need (logically and
physically) to be interconnected by control functionality, but their
media plane functionality may be split. This type of terminals is
referred to as split component terminals. Historically, the earliest
split component terminals were perhaps the (independent) audio and
video conference software tools used over the MBONE in the late
1990s.An example for such a split component terminal is depicted in . Within split component terminal A, at
least audio and video subunits are addressed by their own network
addresses. In some of these systems, the control stack subunit may
also have its own network address.From an RTP viewpoint, each of the subunits terminates RTP, and
acts as an endpoint in the sense that each subunit includes its own,
independent RTP stack. However, as the subunits are semantically part
of the same terminal, it is appropriate that this semantic
relationship is expressed in RTCP protocol elements, namely in the
CNAME.It is further sensible that the subunits share a common clock from
which RTP and RTCP clocks are derived, to facilitate synchronization
and avoid clock drift.To indicate that audio and video Source Streams generated by
different sub-units share a common clock, and can be synchronized, the
RTP streams generated from those Source Streams need to include the
same CNAME in their RTCP SDES packets. The use of a common CNAME for
RTP flows carried in different transport-layer flows is entirely
normal for RTP and RTCP senders, and fully compliant RTP endpoints,
middle-boxes, and other tools should have no problem with this.However, outside of the split component terminal scenario (and
perhaps a multi-homed endpoint scenario, which is not further
discussed herein), the use of a common CNAME in RTP streams sent from
separate endpoints (as opposed to a common CNAME for RTP streams sent
on different transport layer flows between two endpoints) is rare. It
has been reported that at least some third party tools like some
network monitors do not handle endpoints that use of a common CNAME
across multiple transport layer flows gracefully: they report an error
condition that two separate endpoints are using the same CNAME.
Depending on the sophistication of the support staff, such erroneous
reports can lead to support issues.Aforementioned support issue can sometimes be avoided if each of
the subunits of a split component terminal is configured to use a
different CNAME, with the synchronization between the RTP streams
being indicated by some non-RTP signaling channel rather than using a
common CNAME sent in RTCP. This complicates the signaling, especially
in cases where there are multiple SSRCs in use with complex
synchronization requirements, as is the same in many current
telepresence systems. Unless one uses RTCP terminating topologies such
as Topo-RTCP-terminating-MCU, sessions involving more than one video
subunit with a common CNAME are close to unavoidable.The different RTP streams comprising a split terminal system can
form a single RTP session or they can form multiple RTP sessions,
depending on the visibility of their SSRC values in RTCP reports. If
the receiver of the RTP streams sent by the split terminal sends
reports relating to all of the RTP flows (i.e., to each SSRC) in each
RTCP report then a single RTP session is formed. Alternatively, if the
receiver of the RTP streams sent by the split terminal does not send
cross-reports in RTCP, then the audio and video form separate RTP
sessions.For example, in the , B will send
RTCP reports to each of the sub-units of A. If the RTCP packets that B
sends to the audio sub-unit of A include reports on the reception
quality of the video as well as the audio, and similarly if the RTCP
packets that B sends to the video sub-unit of A include reports on the
reception quality of the audio as well as video, then a single RTP
session is formed. However, if the RTCP packets B sends to the audio
sub-unit of A only report on the received audio, and the RTCP packet B
sends to the video sub-unit of A only report on the received video,
then there are two separate RTP sessions.Forming a single RTP session across the RTP streams sent by the
different sub-units of a split terminal gives each sub-unit visibility
into reception quality of RTP streams sent by the other sub-units.
This information can help diagnose reception quality problems, but at
the cost of increased RTCP bandwidth use.RTP streams sent by the sub-units of a split terminal need to use
the same CNAME in their RTCP packets if they are to be synchronized,
irrespective of whether a single RTP session is formed or not.Shortcut name: Topo-AsymmetricIt is theoretically possible to construct an MCU that is a Mixer in
one direction and a Translator in another. The main reason to consider
this would be to allow topologies similar to , where the Mixer does not need to mix in the
direction from B or D towards the multicast domains with A and C.
Instead, the RTP streams from B and D are forwarded without changes.
Avoiding this mixing would save media processing resources that
perform the mixing in cases where it isn't needed. However, there
would still be a need to mix B's media towards D. Only in the
direction B -> multicast domain or D -> multicast domain would
it be possible to work as a Translator. In all other directions, it
would function as a Mixer.The Mixer/Translator would still need to process and change the
RTCP before forwarding it in the directions of B or D to the multicast
domain. One issue is that A and C do not know about the mixed-media
stream the Mixer sends to either B or D. Therefore, any reports
related to these streams must be removed. Also, receiver reports
related to A and C's RTP streams would be missing. To avoid A and C
thinking that B and D aren't receiving A and C at all, the Mixer needs
to insert locally generated reports reflecting the situation for the
streams from A and C into B and D's Sender Reports. In the opposite
direction, the Receiver Reports from A and C about B's and D's stream
also need to be aggregated into the Mixer's Receiver Reports sent to B
and D. Since B and D only have the Mixer as source for the stream, all
RTCP from A and C must be suppressed by the Mixer.This topology is so problematic and it is so easy to get the RTCP
processing wrong, that it is not recommended for implementation.Topologies can be combined and linked to each other using Mixers or
Translators. However, care must be taken in handling the SSRC/CSRC
space. A Mixer does not forward RTCP from sources in other domains,
but instead generates its own RTCP packets for each domain it mixes
into, including the necessary Source Description (SDES) information
for both the CSRCs and the SSRCs. Thus, in a mixed domain, the only
SSRCs seen will be the ones present in the domain, while there can be
CSRCs from all the domains connected together with a combination of
Mixers and Translators. The combined SSRC and CSRC space is common
over any Translator or Mixer. It is important to facilitate loop
detection, something that is likely to be even more important in
combined topologies due to the mixed behavior between the domains. Any
hybrid, like the Topo-Video-switch-MCU or Topo-Asymmetric, requires
considerable thought on how RTCP is dealt with.The topologies discussed in have
different properties. This section describes these properties. Note
that, even if a certain property is supported within a particular
topology concept, the necessary functionality may be optional to
implement.To recapitulate, multicast, and in particular Any Source Multicast
(ASM), provides the functionality that everyone may send to, or
receive from, everyone else within the session. Source-specific
Multicast (SSM) can provide a similar functionality by having anyone
intending to participate as sender to send its media to the SSM
distribution source. The SSM distribution source forwards the media to
all receivers subscribed to the multicast group. Mesh, MCUs, Mixers,
SFMs and Translators may all provide that functionality at least on
some basic level. However, there are some differences in which type of
reachability they provide.Closest to true IP-multicast-based, all-to-all transmission comes
perhaps the transport Translator function called "relay" in , as well as the Mesh with joint RTP
sessions. Media Translators, Mesh with independent RTP Sessions,
Mixers, SFUs and the MCU variants do not provide a fully meshed
forwarding on the transport level; instead, they only allow limited
forwarding of content from the other session participants.The "all to all media transmission" requires that any media
transmitting endpoint considers the path to the least capable
receiving endpoint. Otherwise, the media transmissions may overload
that path. Therefore, a sending endpoint needs to monitor the path
from itself to any of the receiving endpoints, to detect the currently
least capable receiver, and adapt its sending rate accordingly. As
multiple endpoints may send simultaneously, the available resources
may vary. RTCP's Receiver Reports help performing this monitoring, at
least on a medium time scale.The resource consumption for performing all to all transmission
varies depending with the topology. Both ASM and SSM have the benefit
that only one copy of each packet traverses a particular link. Using a
relay causes the transmission of one copy of a packet per
endpoint-to-relay path and packet transmitted. However, in most cases
the links carrying the multiple copies will be the ones close to the
relay (which can be assumed to be part of the network infrastructure
with good connectivity to the backbone), rather than the endpoints
(which may be behind slower access links). The Mesh causes N-1 streams
of transmitted packets to traverse the first hop link from the
endpoint, in an N endpoint mesh. How long the different paths are
common, is highly situation dependent.The transmission of RTCP by design adapts to any changes in the
number of participants due to the transmission algorithm, defined in
the RTP specification, and the
extensions in AVPF (when applicable).
That way, the resources utilized for RTCP stay within the bounds
configured for the session.All Translators, Mixers, and RTCP-terminating MCU, and Mesh with
individual RTP sessions, allow changing the media encoding or the
transport to other properties of the other domain, thereby providing
extended interoperability in cases where the endpoints lack a common
set of media codecs and/or transport protocols. Selective Forwarding
Middleboxes can adopt the transport, and (at least) selectively
forward the encoded streams that match a receiving endpoint's
capability. It requires an additional translator to change the media
encoding if the encoded streams do not match the receiving endpoint's
capabilities.Endpoints are often connected to each other with a heterogeneous
set of paths. This makes congestion control in a Point to Multipoint
set problematic. For the ASM, SSM, Mesh with common RTP session, and
Transport Relay scenario, each individual sending endpoint has to
adapt to the receiving endpoint behind the least capable path,
yielding suboptimal quality for the endpoints behind the more capable
paths. This is no longer an issue when Media Translators, Mixers, SFM
or MCUs are involved, as each endpoint only needs to adapt to the
slowest path within its own domain. The Translator, Mixer, SFM, or MCU
topologies all require their respective outgoing RTP streams to adjust
the bit-rate, packet-rate, etc., to adapt to the least capable path in
each of the other domains. That way one can avoid lowering the quality
to the least-capable endpoint in all the domains at the cost
(complexity, delay, equipment) of the Mixer, SFM or Translator, and
potentially media sender (multicast/layered encoding and sending the
different representations).In the all-to-all media property mentioned above and provided by
ASM, SSM, Mesh with common RTP session, and relay, all simultaneous
media transmissions share the available bit-rate. For endpoints with
limited reception capabilities, this may result in a situation where
even a minimal acceptable media quality cannot be accomplished,
because multiple RTP streams need to share the same resources. One
solution to this problem is to provide for a Mixer, or MCU to
aggregate the multiple RTP streams into a single one, where the single
RTP stream takes up less resources in terms of bit-rate. This
aggregation can be performed according to different methods. Mixing or
selection are two common methods. Selection is almost always possible
and easy to implement. Mixing requires resources in the mixer, and may
be relatively easy and not impairing the quality too badly (audio) or
quite difficult (video tiling, which is not only computationally
complex but also reduces the pixel count per stream, with
corresponding loss in perceptual quality).The RTP protocol includes functionality to identify the session
participants through the use of the SSRC and CSRC fields. In addition,
it is capable of carrying some further identity information about
these participants using the RTCP Source Descriptors (SDES). In
topologies that provide a full all-to-all functionality, i.e. ASM,
Mesh with common RTP session, Relay a compliant RTP implementation
offers the functionality directly as specified in RTP. In topologies
that do not offer all-to-all communication, it is necessary that RTCP
is handled correctly in domain bridging function. RTP includes
explicit specification text for Translators and Mixers, and for SFMs
the required functionality can be derived from that text. However, the
MCU described in cannot offer the
full functionality for session participant identification through RTP
means. The topologies that create independent RTP sessions per
endpoint or pair of endpoints, like Back-to-Back RTP session, MESH
with independent RTP sessions, and the RTCP terminating MCU RTCP terminating MCU, with an exception of
SFM, do not support RTP based identification of session participants.
In all those cases, other non-RTP based mechanisms need to be
implemented if such knowledge is required or desirable. When it comes
to SFM the SSRC name space is not necessarily joint, instead
identification will require knowledge of SSRC/CSRC mappings that the
SFM performed, see .In complex topologies with multiple interconnected domains, it is
possible to unintentionally form media loops. RTP and RTCP support
detecting such loops, as long as the SSRC and CSRC identities are
maintained and correctly set in forwarded packets. Loop detection will
work in ASM, SSM, Mesh with joint RTP session, and Relay. It is likely
that loop detection works for the video switching MCU , at least as long as it forwards the
RTCP between the endpoints. However, the Back-to-Back RTP sessions,
Mesh with independent RTP sessions, SFM, will definitely break the
loop detection mechanism.Some RTP header extensions have relevance not only end-to-end, but
also hop-to-hop, meaning at least some of the middleboxes in the path
are aware of their potential presence through signaling, intercept and
interpret such header extensions and potentially also rewrite or
generate them. Modern header extensions generally follow RFC 5285, which allows for all of the above.
Examples for such header extensions include the mid (media ID) in
[draft-ietf-mmusic-sdp-bundle-negotiation-12]. At the time of
writing there was also a proposal for how to include any SDES into an
RTP header extension
[draft-westerlund-avtext-dses-hdr-ext].When such header extensions are in use, any middlebox that
understands it must ensure consistency between the extensions it sees
and/or generates, and the RTCP it receives and generates. For example,
the mid of bundle is sent in an RTP header extension and also in an
RTCP SDES message. This apparent redundancy was introduced as unaware
middleboxes may choose to discard RTP header extensions. Obviously,
inconsistency between the media ID sent in the RTP header extension
and in the RTCP SDES message could lead to undesirable results, and,
therefore, consistency is needed. Middleboxes unaware of the nature of
a header extension, as specified in RFC
5285, are free to forward or discard header extensions.The table below attempts to summarize the properties of the different
topologies. The legend to the topology abbreviations are:
Topo-Point-to-Point (PtP), Topo-ASM (ASM), Topo-SSM (SSM),
Topo-Trns-Translator (TT), Topo-Media-Translator (including Transport
Translator) (MT), Topo-Mesh with joint session (MJS), Topo-Mesh with
individual sessions (MIS), Topo-Mixer (Mix), Topo-Asymmetric (ASY),
Topo-Video-switch-MCU (VSM), and Topo-RTCP-terminating-MCU (RTM),
Selective Forwarding Middlebox (SFM). In the table below, Y indicates
Yes or full support, N indicates No support, (Y) indicates partial
support, and N/A indicates not applicable.Please note that the Media Translator also includes the transport
Translator functionality.The use of Mixers, SFMs and Translators has impact on security and
the security functions used. The primary issue is that both Mixers, SFMs
and Translators modify packets, thus preventing the use of integrity and
source authentication, unless they are trusted devices that take part in
the security context, e.g., the device can send Secure Realtime Transport Protocol (SRTP) and Secure
Realtime Transport Control Protocol (SRTCP) packets to endpoints
in the Communication Session. If encryption is employed, the media
Translator, SFM and Mixer need to be able to decrypt the media to
perform its function. A transport Translator may be used without access
to the encrypted payload in cases where it translates parts that are not
included in the encryption and integrity protection, for example, IP
address and UDP port numbers in a media stream using SRTP. However, in general, the Translator, SFM
or Mixer needs to be part of the signalling context and get the
necessary security associations (e.g., SRTP crypto contexts) established
with its RTP session participants.Including the Mixer, SFM and Translator in the security context
allows the entity, if subverted or misbehaving, to perform a number of
very serious attacks as it has full access. It can perform all the
attacks possible (see RFC 3550 and any applicable profiles) as if the
media session were not protected at all, while giving the impression to
the human session participants that they are protected.Transport Translators have no interactions with cryptography that
works above the transport layer, such as SRTP, since that sort of
Translator leaves the RTP header and payload unaltered. Media
Translators, on the other hand, have strong interactions with
cryptography, since they alter the RTP payload. A media Translator in a
session that uses cryptographic protection needs to perform
cryptographic processing to both inbound and outbound packets.A media Translator may need to use different cryptographic keys for
the inbound and outbound processing. For SRTP, different keys are
required, because an RFC 3550 media Translator leaves the SSRC unchanged
during its packet processing, and SRTP key sharing is only allowed when
distinct SSRCs can be used to protect distinct packet streams.When the media Translator uses different keys to process inbound and
outbound packets, each session participant needs to be provided with the
appropriate key, depending on whether they are listening to the
Translator or the original source. (Note that there is an architectural
difference between RTP media translation, in which participants can rely
on the RTP Payload Type field of a packet to determine appropriate
processing, and cryptographically protected media translation, in which
participants must use information that is not carried in the
packet.)When using security mechanisms with Translators, SFMs and Mixers, it
is possible that the Translator, SFM or Mixer could create different
security associations for the different domains they are working in.
Doing so has some implications:First, it might weaken security if the Mixer/Translator accepts a
weaker algorithm or key in one domain than in another. Therefore, care
should be taken that appropriately strong security parameters are
negotiated in all domains. In many cases, "appropriate" translates to
"similar" strength. If a key management system does allow the
negotiation of security parameters resulting in a different strength of
the security, then this system should notify the participants in the
other domains about this.Second, the number of crypto contexts (keys and security related
state) needed (for example, in SRTP) may
vary between Mixers, SFMs and Translators. A Mixer normally needs to
represent only a single SSRCs per domain and therefore needs to create
only one security association (SRTP crypto context) per domain. In
contrast, a Translator needs one security association per participant it
translates towards, in the opposite domain. Considering , the Translator needs two
security associations towards the multicast domain, one for B and one
for D. It may be forced to maintain a set of totally independent
security associations between itself and B and D respectively, so as to
avoid two-time pad occurrences. These contexts must also be capable of
handling all the sources present in the other domains. Hence, using
completely independent security associations (for certain keying
mechanisms) may force a Translator to handle N*DM keys and related
state; where N is the total number of SSRCs used over all domains and DM
is the total number of domains.The multicast based (ASM and SSM), Relay and Mesh with common RTP
session are all topologies with multiple endpoints that require shared
knowledge about the different crypto contexts for the endpoints. These
multi-party topologies have special requirements on the key-management
as well as the security functions. Specifically source-authentication in
these environments has special requirements.There exist a number of different mechanisms to provide keys to the
different participants. One example is the choice between group keys and
unique keys per SSRC. The appropriate keying model is impacted by the
topologies one intends to use. The final security properties are
dependent on both the topologies in use and the keying mechanisms'
properties, and need to be considered by the application. Exactly which
mechanisms are used is outside of the scope of this document. Please
review RTP Security Options to get a
better understanding of most of the available options.This document makes no request of IANA.Note to RFC Editor: this section may be removed on publication as an
RFC.The authors would like to thank Mark Baugher, Bo Burman, Umesh
Chandra, Alex Eleftheriadis, Roni Even, Ladan Gharai, Geoff Hunt, Keith
Lantz, Jonathan Lennox, Scarlet Liuyan, Suhas Nandakumar, and Colin
Perkins for their help in reviewing and improving this document.