RFC 3435 - Media Gateway Control Protocol (MGCP) Version 1.0
RFC 3435 - Media Gateway Control Protocol (MGCP) Version 1.0
Or Display the document by number
Network Working Group
F. Andreasen
Request for Comments: 3435
Obsoletes: 2705
Cisco Systems
Category: Informational
January 2003
Media Gateway Control Protocol (MGCP)
Version 1.0
Status of this Memo
This memo provides information for the Internet community.
not specify an Internet standard of any kind.
Distribution of this
memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (2003).
All Rights Reserved.
This document is being published for the information of the
community.
It describes a protocol that is currently being deployed
in a number of products.
Implementers should be aware of ,
which was developed in the IETF Megaco Working Group and the ITU-T
SG16 and which is considered by the IETF and ITU-T to be the
standards-based (including reviewed security considerations) way to
meet the needs that MGCP was designed to address.
This document describes an application programming interface and a
corresponding protocol (MGCP) which is used between elements of a
decomposed multimedia gateway.
The decomposed multimedia gateway
consists of a Call Agent, which contains the call control
"intelligence", and a media gateway which contains the media
functions, e.g., conversion from TDM voice to Voice over IP.
Media gateways contain endpoints on which the Call Agent can create,
modify and delete connections in order to establish and control media
sessions with other multimedia endpoints.
Also, the Call Agent can
instruct the endpoints to detect certain events and generate signals.
The endpoints automatically communicate changes in service state to
the Call Agent.
Furthermore, the Call Agent can audit endpoints as
well as the connections on endpoints.
The basic and general MGCP protocol is defined in this document,
however most media gateways will need to implement one or more MGCP
packages, which define extensions to the protocol suitable for use
with specific types of media gateways.
Such packages are defined in
separate documents.
Table of Contents
Introduction.................................................5
Relation with the H.323 Standards............................7
Relation with the IETF Standards.............................8
Definitions..................................................9
Conventions used in this Document............................9
Media Gateway Control Interface.............................10
Model and Naming Conventions................................10
Types of Endpoints..........................................10
Endpoint Identifiers........................................14
Calls and Connections.......................................16
Names of Call Agents and Other Entities.....................22
Digit Maps..................................................23
Packages....................................................26
Events and Signals..........................................28
Usage of SDP................................................33
Gateway Control Commands....................................33
Overview of Commands........................................33
EndpointConfiguration.......................................36
NotificationRequest.........................................37
Notify......................................................44
CreateConnection............................................46
ModifyConnection............................................52
DeleteConnection (from the Call Agent)......................54
DeleteConnection (from the gateway).........................58
DeleteConnection (multiple connections from the Call Agent) 59
2.3.10 AuditEndpoint...............................................60
2.3.11 AuditConnection.............................................65
2.3.12 RestartInProgress...........................................66
Return Codes and Error Codes................................69
Reason Codes................................................74
Use of Local Connection Options and Connection Descriptors..75
Resource Reservations.......................................77
Media Gateway Control Protocol..............................77
General Description.........................................78
Command Header..............................................79
Command Line................................................79
Parameter Lines.............................................82
Format of response headers.................................101
CreateConnection Response..................................104
ModifyConnection Response..................................105
DeleteConnection Response..................................106
NotificationRequest Response...............................106
Notify Response............................................106
AuditEndpoint Response.....................................106
AuditConnection Response...................................107
RestartInProgress Response.................................108
Encoding of the Session Description (SDP)..................108
Usage of SDP for an Audio Service..........................110
Usage of SDP for LOCAL Connections.........................110
Transmission over UDP......................................111
Providing the At-Most-Once Functionality...................112
Transaction Identifiers and Three Ways Handshake...........113
Computing Retransmission Timers............................114
Maximum Datagram Size, Fragmentation and Reassembly........115
Piggybacking...............................................116
Provisional Responses......................................117
States, Failover and Race Conditions.......................119
Failover Assumptions and Highlights........................119
Communicating with Gateways................................121
Retransmission, and Detection of Lost Associations:........122
Race Conditions............................................126
Quarantine List............................................127
Explicit Detection.........................................133
Transactional Semantics....................................134
Ordering of Commands, and Treatment of Misorder............135
Endpoint Service States....................................137
Fighting the Restart Avalanche.............................140
Disconnected Endpoints.....................................143
Load Control in General....................................146
Security Requirements......................................147
Protection of Media Connections............................148
Packages...................................................148
Actions....................................................150
BearerInformation..........................................150
ConnectionModes............................................151
ConnectionParameters.......................................151
DigitMapLetters............................................151
Events and Signals.........................................152
Default and Reserved Events................................155
ExtensionParameters........................................156
LocalConnectionOptions.....................................157
Reason Codes...............................................157
RestartMethods.............................................158
Return Codes...............................................158
Versions and Compatibility.................................158
Changes from ......................................158
Security Considerations....................................164
Acknowledgments............................................164
References.................................................164
Appendix A: Formal Syntax Description of the Protocol.............167
Appendix B: Base Package..........................................175
Events.....................................................175
Extension Parameters.......................................176
PersistentEvents...........................................176
NotificationState..........................................177
Verbs......................................................177
Appendix C: IANA Considerations...................................179
New MGCP Package Sub-Registry..............................179
New MGCP Package...........................................179
New MGCP LocalConnectionOptions Sub-Registry...............179
Appendix D: Mode Interactions.....................................180
Appendix E: Endpoint Naming Conventions...........................182
Analog Access Line Endpoints...............................182
Digital Trunks.............................................182
Virtual Endpoints..........................................183
Media Gateway..............................................184
Range Wildcards............................................184
Appendix F: Example Command Encodings.............................185
NotificationRequest........................................185
Notify.....................................................186
CreateConnection...........................................186
ModifyConnection...........................................189
DeleteConnection (from the Call Agent).....................189
DeleteConnection (from the gateway)........................190
DeleteConnection (multiple connections
from the Call Agent).......................................190
AuditEndpoint..............................................191
AuditConnection............................................192
RestartInProgress..........................................193
Appendix G: Example Call Flows....................................194
Restart....................................................195
Residential Gateway Restart................................195
Call Agent Restart.........................................198
Connection Creation........................................200
Residential Gateway to Residential Gateway.................200
Connection Deletion........................................206
Residential Gateway to Residential Gateway.................206
Authors' Addresses................................................209
Full Copyright Statement..........................................210
1. Introduction
This document describes an abstract application programming interface
(MGCI) and a corresponding protocol (MGCP) for controlling media
gateways from external call control elements called media gateway
controllers or Call Agents.
A media gateway is typically a network
element that provides conversion between the audio signals carried on
telephone circuits and data packets carried over the Internet or over
other packet networks.
Examples of media gateways are:
* Trunking gateways, that interface between the telephone network and
a Voice over IP network.
Such gateways typically manage a large
number of digital circuits.
* Voice over ATM gateways, which operate much the same way as voice
over IP trunking gateways, except that they interface to an ATM
* Residential gateways, that provide a traditional analog (RJ11)
interface to a Voice over IP network.
Examples of residential
gateways include cable modem/cable set-top boxes, xDSL devices, and
broad-band wireless devices.
* Access gateways, that provide a traditional analog (RJ11) or
digital PBX interface to a Voice over IP network.
Examples of
access gateways include small-scale voice over IP gateways.
* Business gateways, that provide a traditional digital PBX interface
or an integrated "soft PBX" interface to a Voice over IP network.
* Network Access Servers, that can attach a "modem" to a telephone
circuit and provide data access to the Internet.
We expect that in
the future, the same gateways will combine Voice over IP services
and Network Access services.
* Circuit switches, or packet switches, which can offer a control
interface to an external call control element.
MGCP assumes a call control architecture where the call control
"intelligence" is outside the gateways and handled by external call
control elements known as Call Agents.
The MGCP assumes that these
call control elements, or Call Agents, will synchronize with each
other to send coherent commands and responses to the gateways under
their control.
If this assumption is violated, inconsistent behavior
should be expected.
MGCP does not define a mechanism for
synchronizing Call Agents.
MGCP is, in essence, a master/slave
protocol, where the gateways are expected to execute commands sent by
the Call Agents.
In consequence, this document specifies in great
detail the expected behavior of the gateways, but only specifies
those parts of a Call Agent implementation, such as timer management,
that are mandated for proper operation of the protocol.
MGCP assumes a connection model where the basic constructs are
endpoints and connections.
Endpoints are sources and/or sinks of
data and can be physical or virtual.
Examples of physical endpoints
* An interface on a gateway that terminates a trunk connected to a
PSTN switch (e.g., Class 5, Class 4, etc.).
A gateway that
terminates trunks is called a trunking gateway.
* An interface on a gateway that terminates an analog POTS connection
to a phone, key system, PBX, etc.
A gateway that terminates
residential POTS lines (to phones) is called a residential gateway.
An example of a virtual endpoint is an audio source in an audio-
content server.
Creation of physical endpoints requires hardware
installation, while creation of virtual endpoints can be done by
Connections may be either point to point or multipoint.
A point to
point connection is an association between two endpoints with the
purpose of transmitting data between these endpoints.
association is established for both endpoints, data transfer between
these endpoints can take place.
A multipoint connection is
established by connecting the endpoint to a multipoint session.
Connections can be established over several types of bearer networks,
for example:
* Transmission of audio packets using RTP and UDP over an IP network.
* Transmission of audio packets using AAL2, or another adaptation
layer, over an ATM network.
* Transmission of packets over an internal connection, for example
the TDM backplane or the interconnection bus of a gateway.
used, in particular, for "hairpin" connections, connections that
terminate in a gateway but are immediately rerouted over the
telephone network.
For point-to-point connections the endpoints of a connection could be
in separate gateways or in the same gateway.
1.1 Relation with the H.323 Standards
MGCP is designed as an internal protocol within a distributed system
that appears to the outside as a single VoIP gateway.
This system is
composed of a Call Agent, that may or may not be distributed over
several computer platforms, and of a set of gateways, including at
least one "media gateway" that perform the conversion of media
signals between circuits and packets, and at least one "signaling
gateway" when connecting to an SS7 controlled network.
In a typical
configuration, this distributed gateway system will interface on one
side with one or more telephony (i.e., circuit) switches, and on the
other side with H.323 conformant systems, as indicated in the
following table:
------------------------------------------------------------------
| Functional|
Terminating
H.323 conformant
|-----------|------------|-----------------|-----------------------|
| Signaling |
Signaling |
Call agent
Signaling exchanges
exchanges |
with the Call Agent
through H.225/RAS and|
H.225/Q.931.
|-----------|------------|-----------------|-----------------------|
Possible negotiation |
of logical channels
and transmission
parameters through
H.245 with the call
|-----------|------------|-----------------|-----------------------|
synchronization|
through MGCP
|-----------|------------|-----------------|-----------------------|
Connection|
Transmission of VoIP |
data using RTP
| Transport |
high speed|
directly between the |
H.323 station and the|
------------------------------------------------------------------
In the MGCP model, the gateways focus on the audio signal translation
function, while the Call Agent handles the call signaling and call
processing functions.
As a consequence, the Call Agent implements
the "signaling" layers of the H.323 standard, and presents itself as
an "H.323 Gatekeeper" or as one or more "H.323 Endpoints" to the
H.323 systems.
Relation with the IETF Standards
While H.323 is the recognized standard for VoIP terminals, the IETF
has also produced specifications for other types of multi-media
applications.
These other specifications include:
* the Session Description Protocol (SDP),
* the Session Announcement Protocol (SAP),
* the Session Initiation Protocol (SIP),
* the Real Time Streaming Protocol (RTSP), .
The latter three specifications are in fact alternative signaling
standards that allow for the transmission of a session description to
an interested party.
SAP is used by multicast session managers to
distribute a multicast session description to a large group of
recipients, SIP is used to invite an individual user to take part in
a point-to-point or unicast session, RTSP is used to interface a
server that provides real time data.
In all three cases, the session
description is described according to SDP; when audio is transmitted,
it is transmitted through the Real-time Transport Protocol, RTP.
The distributed gateway systems and MGCP will enable PSTN telephony
users to access sessions set up using SAP, SIP or RTSP.
Agent provides for signaling conversion, according to the following
------------------------------------------------------------------
| Functional|
Terminating
IETF conforming systems|
|-----------|------------|---------------|-------------------------|
| Signaling |
Signaling |
Call agent
Signaling exchanges
exchanges |
with the Call Agent
through SAP, SIP or
|-----------|------------|---------------|-------------------------|
Negotiation of session |
description parameters |
through SDP (telephony |
gateway terminated but |
passed via the call
agent to and from the
IETF conforming system)|
|-----------|------------|---------------|-------------------------|
| Internal syn- |
| chronization
| through MGCP
|-----------|------------|---------------|-------------------------|
Connection|
Transmission of VoIP
data using RTP,
| Transport |
high speed|
directly between the
remote IP end system
and the gateway.
------------------------------------------------------------------
The SDP standard has a pivotal status in this architecture.
see in the following description that we also use it to carry session
descriptions in MGCP.
1.3 Definitions
A communication channel between two switching systems, e.g.,
a DS0 on a T1 or E1 line.
1.4 Conventions used in this Document
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED, "MAY", and
"OPTIONAL" in this document are to be interpreted as described in BCP
2. Media Gateway Control Interface
The interface functions provide for connection control and endpoint
Both use the same system model and the same naming
conventions.
2.1 Model and Naming Conventions
The MGCP assumes a connection model where the basic constructs are
endpoints and connections.
Connections are grouped in calls.
more connections can belong to one call.
Connections and calls are
set up at the initiative of one or more Call Agents.
2.1.1 Types of Endpoints
In the introduction, we presented several classes of gateways.
classifications, however, can be misleading.
Manufacturers can
arbitrarily decide to provide several types of services in a single
A single product could well, for example, provide some
trunk connections to telephony switches, some primary rate
connections and some analog line interfaces, thus sharing the
characteristics of what we described in the introduction as
"trunking", "access" and "residential" gateways.
MGCP does not make
assumptions about such groupings.
We simply assume that media
gateways support collections of endpoints.
The type of the endpoint
determines its functionality.
Our analysis, so far, has led us to
isolate the following basic endpoint types:
* Digital channel (DS0),
* Analog line,
* Announcement server access point,
* Interactive Voice Response access point,
* Conference bridge access point,
* Packet relay,
* ATM "trunk side" interface.
In this section, we will describe the expected behavior of such
endpoints.
This list is not final.
There may be other types of endpoints
defined in the future, for example test endpoints that could be used
to check network quality, or frame-relay endpoints that could be used
to manage audio channels multiplexed over a frame-relay virtual
2.1.1.1 Digital Channel (DS0)
Digital channels provide a 64 Kbps service.
Such channels are found
in trunk and ISDN interfaces.
They are typically part of digital
multiplexes, such as T1, E1, T3 or E3 interfaces.
Media gateways
that support such channels are capable of translating the digital
signals received on the channel, which may be encoded according to
A-law or mu-law, using either the complete set of 8 bits per sample
or only 7 of these bits, into audio packets.
When the media gateway
also supports a Network Access Server (NAS) service, the gateway
shall be capable of receiving either audio-encoded data (modem
connection) or binary data (ISDN connection) and convert them into
data packets.
+------------+|
(channel) ===|DS0 endpoint| -------- Connections
+------------+|
Media gateways should be able to establish several connections
between the endpoint and the packet networks, or between the endpoint
and other endpoints in the same gateway.
The signals originating
from these connections shall be mixed according to the connection
"mode", as specified later in this document.
The precise number of
connections that an endpoint supports is a characteristic of the
gateway, and may in fact vary according to the allocation of
resources within the gateway.
In some cases, digital channels are used to carry signaling.
the case for example for SS7 "F" links, or ISDN "D" channels.
gateways that support these signaling functions shall be able to send
and receive the signaling packets to and from a Call Agent, using the
"backhaul" procedures defined by the SIGTRAN working group of the
Digital channels are sometimes used in conjunction with
channel associated signaling, such as "MF R2".
Media gateways that
support these signaling functions shall be able to detect and produce
the corresponding signals, such as for example "wink" or "A",
according to the event signaling and reporting procedures defined in
2.1.1.2 Analog Line
Analog lines can be used either as a "client" interface, providing
service to a classic telephone unit, or as a "service" interface,
allowing the gateway to send and receive analog calls.
media gateway also supports a NAS service, the gateway shall be
capable of receiving audio-encoded data (modem connection) and
convert them into data packets.
+---------------+|
(line) ===|analog endpoint| -------- Connections
+---------------+|
Media gateways should be able to establish several connections
between the endpoint and the packet networks, or between the endpoint
and other endpoints in the same gateway.
The audio signals
originating from these connections shall be mixed according to the
connection "mode", as specified later in this document.
The precise
number of connections that an endpoint supports is a characteristic
of the gateway, and may in fact vary according to the allocation of
resources within the gateway.
A typical gateway should however be
able to support two or three connections per endpoint, in order to
support services such as "call waiting" or "three way calling".
2.1.1.3 Announcement Server Access Point
An announcement server endpoint provides access to an announcement
Under requests from the Call Agent, the announcement server
will "play" a specified announcement.
The requests from the Call
Agent will follow the event signaling and reporting procedures
defined in MGCP.
+----------------------+
| Announcement endpoint| -------- Connection
+----------------------+
A given announcement endpoint is not expected to support more than
one connection at a time.
If several connections were established to
the same endpoint, then the same announcements would be played
simultaneously over all the connections.
Connections to an announcement server are typically one way, or "half
duplex" -- the announcement server is not expected to listen to the
audio signals from the connection.
2.1.1.4 Interactive Voice Response Access Point
An Interactive Voice Response (IVR) endpoint provides access to an
IVR service.
Under requests from the Call Agent, the IVR server will
"play" announcements and tones, and will "listen" to responses, such
as DTMF input or voice messages, from the user.
The requests from
the Call Agent will follow the event signaling and reporting
procedures defined in MGCP.
+-------------+
| IVR endpoint| -------- Connection
+-------------+
A given IVR endpoint is not expected to support more than one
connection at a time.
If several connections were established to the
same endpoint, then the same tones and announcements would be played
simultaneously over all the connections.
2.1.1.5 Conference Bridge Access Point
A conference bridge endpoint is used to provide access to a specific
conference.
+--------------------------+|
|Conference bridge endpoint| -------- Connections
+--------------------------+|
Media gateways should be able to establish several connections
between the endpoint and the packet networks, or between the endpoint
and other endpoints in the same gateway.
The signals originating
from these connections shall be mixed according to the connection
"mode", as specified later in this document.
The precise number of
connections that an endpoint supports is a characteristic of the
gateway, and may in fact vary according to the allocation of
resources within the gateway.
2.1.1.6 Packet Relay
A packet relay endpoint is a specific form of conference bridge, that
typically only supports two connections.
Packets relays can be found
in firewalls between a protected and an open network, or in
transcoding servers used to provide interoperation between
incompatible gateways, for example gateways that do not support
compatible compression algorithms, or gateways that operate over
different transmission networks such as IP and ATM.
+---------------------+ |
|Packet relay endpoint|
2 connections
+---------------------+ |
2.1.1.7 ATM "trunk side" Interface
ATM "trunk side" endpoints are typically found when one or several
ATM permanent virtual circuits are used as a replacement for the
classic "TDM" trunks linking switches.
When ATM/AAL2 is used,
several trunks or channels are multiplexed on a single virtual
each of these trunks correspond to a single endpoint.
+------------------+|
(channel) = |ATM trunk endpoint| -------- Connections
+------------------+|
Media gateways should be able to establish several connections
between the endpoint and the packet networks, or between the endpoint
and other endpoints in the same gateway.
The signals originating
from these connections shall be mixed according to the connection
"mode", as specified later in this document.
The precise number of
connections that an endpoint supports is a characteristic of the
gateway, and may in fact vary according to the allocation of
resources within the gateway.
2.1.2 Endpoint Identifiers
Endpoint identifiers have two components that both are case-
insensitive:
* the domain name of the gateway that is managing the endpoint
* a local name within that gateway
Endpoint names are of the form:
local-endpoint-name@domain-name
where domain-name is an absolute domain-name as defined in
and includes a host portion, thus an example domain-name could be:
mygateway.whatever.net
Also, domain-name may be an IP-address of the form defined for domain
name in , thus another example could be (see
[192.168.1.2]
Both IPv4 and IPv6 addresses can be specified, however use of IP
addresses as endpoint identifiers is generally discouraged.
Note that since the domain name portion is part of the endpoint
identifier, different forms or different values referring to the same
entity are not freely interchangeable.
The most recently supplied
form and value MUST always be used.
The local endpoint name is case-insensitive.
The syntax of the local
endpoint name is hierarchical, where the least specific component of
the name is the leftmost term, and the most specific component is the
rightmost term.
The precise syntax depends on the type of endpoint
being named and MAY start with a term that identifies the endpoint
In any case, the local endpoint name MUST adhere to the
following naming rules:
1) The individual terms of the naming path MUST be separated by a
single slash ("/", ASCII 2F hex).
2) The individual terms are character strings composed of letters,
digits or other printable characters, with the exception of
characters used as delimiters ("/", "@"), characters used for
wildcarding ("*", "$") and white spaces.
3) Wild-carding is represented either by an asterisk ("*") or a
dollar sign ("$") for the terms of the naming path which are to be
wild-carded.
Thus, if the full local endpoint name is of the
term1/term2/term3
then the entity name field looks like this depending on which
terms are wild-carded:
*/term2/term3 if term1 is wild-carded
term1/*/term3 if term2 is wild-carded
term1/term2/* if term3 is wild-carded
if term2 and term3 are wild-carded, etc.
In each of these examples a dollar sign could have appeared
instead of an asterisk.
4) A term represented by an asterisk ("*") is to be interpreted as:
"use ALL values of this term known within the scope of the Media
Unless specified otherwise, this refers to all
endpoints configured for service, regardless of their actual
service state, i.e., in-service or out-of-service.
5) A term represented by a dollar sign ("$") is to be interpreted as:
"use ANY ONE value of this term known within the scope of the
Media Gateway".
Unless specified otherwise, this only refers to
endpoints that are in-service.
Furthermore, it is RECOMMENDED that Call Agents adhere to the
following:
* Wild-carding should only be done from the right, thus if a term is
wild-carded, then all terms to the right of that term should be
wild-carded as well.
* In cases where mixed dollar sign and asterisk wild-cards are used,
dollar-signs should only be used from the right, thus if a term had
a dollar sign wild-card, all terms to the right of that term should
also contain dollar sign wild-cards.
The description of a specific command may add further criteria for
selection within the general rules given above.
Note, that wild-cards may be applied to more than one term in which
case they shall be evaluated from left to right.
For example, if we
have the endpoint names "a/1", "a/2", "b/1", and "b/2", then "$/*"
(which is not recommended) will evaluate to either "a/1, a/2", or
"b/1, b/2".
However, "*/$" may evaluate to "a/1, b/1", "a/1, b/2",
"a/2, b/1", or "a/2, b/2".
The use of mixed wild-cards in a command
is considered error prone and is consequently discouraged.
A local name that is composed of only a wildcard character refers to
either all (*) or any ($) endpoints within the media gateway.
2.1.3 Calls and Connections
Connections are created on the Call Agent on each endpoint that will
be involved in the "call".
In the classic example of a connection
between two "DS0" endpoints (EP1 and EP2), the Call Agents
controlling the endpoints will establish two connections (C1 and C2):
(channel1) ===|EP1|--(C1)--...
...(C2)--|EP2|===(channel2)
Each connection will be designated locally by an endpoint unique
connection identifier, and will be characterized by connection
attributes.
When the two endpoints are located on gateways that are managed by
the same Call Agent, the creation is done via the three following
1) The Call Agent asks the first gateway to "create a connection" on
the first endpoint.
The gateway allocates resources to that
connection, and responds to the command by providing a "session
description".
The session description contains the information
necessary for a third party to send packets towards the newly
created connection, such as for example IP address, UDP port, and
codec parameters.
2) The Call Agent then asks the second gateway to "create a
connection" on the second endpoint.
The command carries the
"session description" provided by the first gateway.
The gateway
allocates resources to that connection, and responds to the
command by providing its own "session description".
3) The Call Agent then uses a "modify connection" command to provide
this second "session description" to the first endpoint.
this is done, communication can proceed in both directions.
When the two endpoints are located on gateways that are managed by
two different Call Agents, the Call Agents exchange information
through a Call-Agent to Call-Agent signaling protocol, e.g., SIP [7],
in order to synchronize the creation of the connection on the two
endpoints.
Once a connection has been established, the connection parameters can
be modified at any time by a "modify connection" command.
Agent may for example instruct the gateway to change the codec used
on a connection, or to modify the IP address and UDP port to which
data should be sent, if a connection is "redirected".
The Call Agent removes a connection by sending a "delete connection"
command to the gateway.
The gateway may also, under some
circumstances, inform a gateway that a connection could not be
sustained.
The following diagram provides a view of the states of a connection,
as seen from the gateway:
Create connection
+-------------------+
|resource allocation|-(failed)-+
+-------------------+
(connection refused)
(successful)
+-----------&+
+-------------------+
remote session
description
|----------(yes)--------+
available ?
+-------------------+
+-----------+
| +---&| half open |------& Delete
&-------| open |&----------+
Connection
+-----------+
Modify Connection
Modify Connection
| | +--------------------+
+--------------------+
| | |assess modification |
|assess modification |
| | +--------------------+
+--------------------+
| |(failed)
(successful)
(successful) |
+-------------+-------+
+&-------------------+
+-----------------+
| Free connection |
| resources.
+-----------------+
2.1.3.1 Names of Calls
One of the attributes of each connection is the "call identifier",
which as far as the MGCP protocol is concerned has little semantic
meaning, and is mainly retained for backwards compatibility.
Calls are identified by unique identifiers, independent of the
underlying platforms or agents.
Call identifiers are hexadecimal
strings, which are created by the Call Agent.
The maximum length of
call identifiers is 32 characters.
Call identifiers are expected to be unique within the system, or at a
minimum, unique within the collection of Call Agents that control the
same gateways.
From the gateway's perspective, the Call identifier
is thus unique.
When a Call Agent builds several connections that
pertain to the same call, either on the same gateway or in different
gateways, these connections that belong to the same call should share
the same call-id.
This identifier can then be used by accounting or
management procedures, which are outside the scope of MGCP.
2.1.3.2 Names of Connections
Connection identifiers are created by the gateway when it is
requested to create a connection.
They identify the connection
within the context of an endpoint.
Connection identifiers are
treated in MGCP as hexadecimal strings.
The gateway MUST make sure
that a proper waiting period, at least 3 minutes, elapses between the
end of a connection that used this identifier and its use in a new
connection for the same endpoint (gateways MAY decide to use
identifiers that are unique within the context of the gateway).
maximum length of a connection identifier is 32 characters.
2.1.3.3 Management of Resources, Attributes of Connections
Many types of resources will be associated to a connection, such as
specific signal processing functions or packetization functions.
Generally, these resources fall in two categories:
1) Externally visible resources, that affect the format of "the bits
on the network" and must be communicated to the second endpoint
involved in the connection.
2) Internal resources, that determine which signal is being sent over
the connection and how the received signals are processed by the
The resources allocated to a connection, and more generally the
handling of the connection, are chosen by the gateway under
instructions from the Call Agent.
The Call Agent will provide these
instructions by sending two sets of parameters to the gateway:
1) The local directives instruct the gateway on the choice of
resources that should be used for a connection,
2) When available, the "session description" provided by the other
end of the connection (referred to as the remote session
description).
The local directives specify such parameters as the mode of the
connection (e.g., send-only, or send-receive), preferred coding or
packetization methods, usage of echo cancellation or silence
suppression.
(A detailed list can be found in the specification of
the LocalConnectionOptions parameter of the CreateConnection
Depending on the parameter, the Call Agent MAY either
specify a value, a range of values, or no value at all.
This allows
various implementations to implement various levels of control, from
a very tight control where the Call Agent specifies minute details of
the connection handling to a very loose control where the Call Agent
only specifies broad guidelines, such as the maximum bandwidth, and
lets the gateway choose the detailed values subject to the
guidelines.
Based on the value of the local directives, the gateway will
determine the resources to allocate to the connection.
When this is
possible, the gateway will choose values that are in line with the
remote session description - but there is no absolute requirement
that the parameters be exactly the same.
Once the resources have been allocated, the gateway will compose a
"session description" that describes the way it intends to send and
receive packets.
Note that the session description may in some cases
present a range of values.
For example, if the gateway is ready to
accept one of several compression algorithms, it can provide a list
of these accepted algorithms.
Local Directives
(from Call Agent 1)
+-------------+
| resource
| allocation
| (gateway 1) |
+-------------+
Parameters
Description
Local Directives
(from Call Agent 2)
+---& Transmission----+
(CA to CA)
+-------------+
| resource
| allocation
| (gateway 2) |
+-------------+
Parameters
Description
+---- Transmission&---+
(CA to CA)
+-------------+
| modification|
| (gateway 1) |
+-------------+
Parameters
-- Information flow: local directives & session descriptions --
2.1.3.4 Special Case of Local Connections
Large gateways include a large number of endpoints which are often of
different types.
In some networks, we may often have to set-up
connections between endpoints that are located within the same
Examples of such connections may be:
* Connecting a call to an Interactive Voice-Response unit,
* Connecting a call to a Conferencing unit,
* Routing a call from one endpoint to another, something often
described as a "hairpin" connection.
Local connections are much simpler to establish than network
connections.
In most cases, the connection will be established
through some local interconnecting device, such as for example a TDM
When two endpoints are managed by the same gateway, it is possible to
specify the connection in a single command that conveys the names of
the two endpoints that will be connected.
The command is essentially
a "Create Connection" command which includes the name of the second
endpoint in lieu of the "remote session description".
2.1.4 Names of Call Agents and Other Entities
The media gateway control protocol has been designed to allow the
implementation of redundant Call Agents, for enhanced network
reliability.
This means that there is no fixed binding between
entities and hardware platforms or network interfaces.
Call Agent names consist of two parts, similar to endpoint names.
Semantically, the local portion of the name does not exhibit any
internal structure.
An example Call Agent name is:
Note that both the local part and the domain name have to be
supplied. Nevertheless, implementations are encouraged to accept call
agent names consisting of only the domain name.
Reliability can be improved by using the following procedures:
* Entities such as endpoints or Call Agents are identified by their
domain name, not their network addresses.
Several addresses can be
associated with a domain name.
If a command or a response cannot
be forwarded to one of the network addresses, implementations MUST
retry the transmission using another address.
* Entities MAY move to another platform.
The association between a
logical name (domain name) and the actual platform is kept in the
domain name service.
Call Agents and Gateways MUST keep track of
the time-to-live of the record they read from the DNS.
query the DNS to refresh the information if the time to live has
In addition to the indirection provided by the use of domain names
and the DNS, the concept of "notified entity" is central to
reliability and fail-over in MGCP.
The "notified entity" for an
endpoint is the Call Agent currently controlling that endpoint.
any point in time, an endpoint has one, and only one, "notified
entity" associated with it.
The "notified entity" determines where
the endpoint w when the endpoint needs to send a
command to the Call Agent, it MUST send the command to its current
"notified entity".
The "notified entity" however does not determine
where commands
any Call Agent can send commands
to the endpoint.
Please refer to Section 5 for the relevant security
considerations.
Upon startup, the "notified entity" MUST be set to a provisioned
Most commands sent by the Call Agent include the ability to
explicitly name the "notified entity" through the use of a
"NotifiedEntity" parameter.
The "notified entity" will stay the same
until either a new "NotifiedEntity" parameter is received or the
endpoint does a warm or cold (power-cycle) restart.
If a "NotifiedEntity" parameter is sent with an "empty" value, the
"notified entity" for the endpoint will be set to empty.
"notified entity" for an endpoint is empty or has not been set
explicitly (neither by a command nor by provisioning), the "notified
entity" will then default to the source address (i.e., IP address and
UDP port number) of the last successful non-audit command received
for the endpoint.
Auditing will thus not change the "notified
Use of an empty "NotifiedEntity" parameter value is
strongly discouraged as it is error prone and eliminates the DNS-
based fail-over and reliability mechanisms.
2.1.5 Digit Maps
The Call Agent can ask the gateway to collect digits dialed by the
This facility is intended to be used with residential gateways
to collect the numbe it can also be used with
trunking gateways and access gateways alike, to collect access codes,
credit card numbers and other numbers requested by call control
One procedure is for the gateway to notify the Call Agent of each
individual dialed digit, as soon as they are dialed.
However, such a
procedure generates a large number of interactions.
It is preferable
to accumulate the dialed numbers in a buffer, and to transmit them in
a single message.
The problem with this accumulation approach, however, is that it is
hard for the gateway to predict how many numbers it needs to
accumulate before transmission.
For example, using the phone on our
desk, we can dial the following numbers:
------------------------------------------------------
Local operator
Long distance operator
Local extension number
Local number
Shortcut to local number at|
other corporate sites
Star services
91xxxxxxxxxx
Long distance number
9011 + up to 15 digits|
International number
------------------------------------------------------
The solution to this problem is to have the Call Agent load the
gateway with a digit map that may correspond to the dial plan.
digit map is expressed using a syntax derived from the Unix system
command, egrep.
For example, the dial plan described above results
in the following digit map:
(0T|00T|[1-7]xxx|8xxxxxxx|#xxxxxxx|*xx|91xxxxxxxxxx|9011x.T)
The formal syntax of the digit map is described by the DigitMap rule
in the formal syntax description of the protocol (see Appendix A) -
support for basic digit map letters is REQUIRED while support for
extension digit map letters is OPTIONAL.
A gateway receiving a digit
map with an extension digit map letter not supported SHOULD return
error code 537 (unknown digit map extension).
A digit map, according to this syntax, is defined either by a (case
insensitive) "string" or by a list of strings.
Each string in the
list is an alternative numbering scheme, specified either as a set of
digits or timers, or as an expression over which the gateway will
attempt to find a shortest possible match.
The following constructs
can be used in each numbering scheme:
A digit from "0" to "9".
The symbol "T" matching a timer expiry.
A digit, a timer, or one of the symbols "A", "B", "C",
"D", "#", or "*".
Extensions may be defined.
* Wildcard: The symbol "x" which matches any digit ("0" to "9").
One or more DTMF symbols enclosed between square brackets
("[" and "]").
* Subrange: Two digits separated by hyphen ("-") which matches any
digit between and including the two.
The subrange
construct can only be used inside a range construct,
i.e., between "[" and "]".
* Position: A period (".") which matches an arbitrary number,
including zero, of occurrences of the preceding
construct.
A gateway that detects events to be matched against a digit map MUST
do the following:
1) Add the event code as a token to the end of an internal state
variable for the endpoint called the "current dial string".
2) Apply the current dial string to the digit map table, attempting a
match to each expression in the digit map.
3) If the result is under-qualified (partially matches at least one
entry in the digit map and doesn't completely match another
entry), do nothing further.
If the result matches an entry, or is over-qualified (i.e., no
further digits could possibly produce a match), send the list of
accumulated events to the Call Agent.
A match, in this
specification, can be either a "perfect match," exactly matching one
of the specified alternatives, or an impossible match, which occurs
when the dial string does not match any of the alternatives.
Unexpected timers, for example, can cause "impossible matches".
perfect matches and impossible matches trigger notification of the
accumulated digits (which may include other events - see Section
The following example illustrates the above.
Assume we have the
digit map:
(xxxxxxx|x11)
and a current dial string of "41".
Given the input "1" the current
dial string becomes "411".
We have a partial match with "xxxxxxx",
but a complete match with "x11", and hence we send "411" to the Call
The following digit map example is more subtle:
(0[12].|00|1[12].1|2x.#)
Given the input "0", a match will occur immediately since position
(".") allows for zero occurrences of the preceding construct.
input "00" can thus never be produced in this digit map.
Given the input "1", only a partial match exists.
The input "12" is
also only a partial match, however both "11" and "121" are a match.
Given the input "2", a partial match exists.
A partial match also
exists for the input "23", "234", "2345", etc.
A full match does not
occur here until a "#" is generated, e.g., "2345#".
The input "2#"
would also have been a match.
Note that digit maps simply define a way of matching sequences of
event codes against a grammar.
Although digit maps as defined here
are for DTMF input, extension packages can also be defined so that
digit maps can be used for other types of input represented by event
codes that adhere to the digit map syntax already defined for these
event codes (e.g., "1" or "T").
Where such usage is envisioned, the
definition of the particular event(s) SHOULD explicitly state that in
the package definition.
Since digit maps are not bounded in size, it is RECOMMENDED that
gateways support digit maps up to at least 2048 bytes per endpoint.
2.1.6 Packages
MGCP is a modular and extensible protocol, however with extensibility
comes the need to manage, identify, and name the individual
extensions.
This is achieved by the concept of packages, which are
simply well-defined groupings of extensions.
For example, one
package may support a certain group of events and signals, e.g.,
off-hook and ringing, for analog access lines.
Another package may
support another group of events and signals for analog access lines
or for another type of endpoint such as video.
One or more packages
may be supported by a given endpoint.
MGCP allows the following types of extensions to be defined in a
* BearerInformation
* LocalConnectionOptions
* ExtensionParameters
* ConnectionModes
* DigitMapLetters
* ConnectionParameters
* RestartMethods
* ReasonCodes
* Return codes
each of which will be explained in more detail below.
The rules for
defining each of these extensions in a package are described in
Section 6, and the encoding and syntax are defined in Section 3 and
Appendix A.
With the exception of DigitMapLetters, a package defines a separate
name space for each type of extension by adding the package name as a
prefix to the extension, i.e.:
package-name/extension
Thus the package-name is followed by a slash ("/") and the name of
the extension.
An endpoint supporting one or more packages may define one of those
packages as the default package for the endpoint.
Use of the package
name for events and signals in the default package for an endpoint is
OPTIONAL, however it is RECOMMENDED to always include the package
All other extensions, except DigitMapLetter, defined in the
package MUST include the package-name when referring to the
extension.
Package names are case insensitive strings of letters, hyphens and
digits, with the restriction that hyphens shall never be the first or
last character in a name.
Examples of package names are "D", "T",
and "XYZ".
Package names are not case sensitive - names such as
"XYZ", "xyz", and "xYz" are equal.
Package definitions will be provided in other documents and with
package names and extensions names registered with IANA.
details, refer to section 6.
Implementers can gain experience by using experimental packages.
name of an experimental package MUST start with the two characters
"x-"; the IANA SHALL NOT register package names that start with these
characters, or the characters "x+", which are reserved.
that receives a command referring to an unsupported package MUST
return an error (error code 518 - unsupported package, is
RECOMMENDED).
2.1.7 Events and Signals
The concept of events and signals is central to MGCP.
A Call Agent
may ask to be notified about certain events occurring in an endpoint
(e.g., off-hook events) by including the name of the event in a
RequestedEvents parameter (in a NotificationRequest command - see
Section 2.3.3).
A Call Agent may also request certain signals to be applied to an
endpoint (e.g., dial-tone) by supplying the name of the event in a
SignalRequests parameter.
Events and signals are grouped in packages, within which they share
the same name space which we will refer to as event names in the
following.
Event names are case insensitive strings of letters,
hyphens and digits, with the restriction that hyphens SHALL NOT be
the first or last character in a name.
Some event codes may need to
be parameterized with additional data, which is accomplished by
adding the parameters between a set of parentheses.
Event names are
not case sensitive - values such as "hu", "Hu", "HU" or "hU" are
Examples of event names can be "hu" (off hook or "hang-up"
transition), "hf" (hook-flash) or "0" (the digit zero).
The package name is OPTIONAL for events in the default package for an
endpoint, however it is RECOMMENDED to always include the package
If the package name is excluded from the event name, the
default package name for that endpoint MUST be assumed.
For example,
for an analog access line which has the line package ("L") as a
default with dial-tone ("dl") as one of the events in that package,
the following two event names are equal:
For any other non-default packages that are associated with that
endpoint, (such as the generic package for an analog access
endpoint-type for example), the package name MUST be included with
the event name.
Again, unconditional inclusion of the package name
is RECOMMENDED.
Digits, or letters, are supported in some packages, notably "DTMF".
Digits and letters are defined by the rules "Digit" and "Letter" in
the definition of digit maps.
This definition refers to the digits
(0 to 9), to the asterisk or star ("*") and orthotrope, number or
pound sign ("#"), and to the letters "A", "B", "C" and "D", as well
as the timer indication "T".
These letters can be combined in "digit
string" that represents the keys that a user punched on a dial.
addition, the letter "X" can be used to represent all digits (0 to
Also, extensions MAY define use of other letters.
The need to
easily express the digit strings in earlier versions of the protocol
has a consequence on the form of event names:
An event name that does not denote a digit MUST always contain at
least one character that is neither a digit, nor one of the letters
A, B, C, D, T or X (such names also MUST NOT just contain the special
signs "*", or "#").
Event names consisting of more than one
character however may use any of the above.
A Call Agent may often have to ask a gateway to detect a group of
Two conventions can be used to denote such groups:
* The "*" and "all" wildcard conventions (see below) can be used to
detect any event belonging to a package, or a given event in many
packages, or any event in any package supported by the gateway.
* The regular expression Range notation can be used to detect a range
of digits.
The star sign (*) can be used as a wildcard instead of a package
name, and the keyword "all" can be used as a wildcard instead of an
event name:
* A name such as "foo/all" denotes all events in package "foo".
* A name such as "*/bar" denotes the event "bar" in any package
supported by the gateway.
* The name "*/all" denotes all events supported by the endpoint.
This specification purposely does not define any additional detail
for the "all packages" and "all events" wildcards.
They provide
limited benefits, but introduce significant complexity along with the
potential for errors.
Their use is consequently strongly
discouraged.
The Call Agent can ask a gateway to detect a set of digits or letters
either by individually describing those letters, or by using the
"range" notation defined in the syntax of digit strings.
example, the Call Agent can:
* Use the letter "x" to denote" digits from 0 to 9.
* Use the notation "[0-9#]" to denote the digits 0 to 9 and the pound
The individual event codes are still defined in a package though
(e.g., the "DTMF" package).
Events can by default only be generated and detected on endpoints,
however events can be also be defined so they can be generated or
detected on connections rather than on the endpoint itself (see
Section 6.6).
For example, gateways may be asked to provide a
ringback tone on a connection.
When an event is to be applied on a
connection, the name of the connection MUST be added to the name of
the event, using an "at" sign (@) as a delimiter, as in:
G/rt@0A3F58
where "G" is the name of the package and "rt" is the name of the
Should the connection be deleted while an event or signal is
being detected or applied on it, that particular event detection or
signal generation simply stops.
Depending on the signal, this may
generate a failure (see below).
The wildcard character "*" (star) can be used to denote "all
connections".
When this convention is used, the gateway will
generate or detect the event on all the connections that are
connected to the endpoint.
This applies to existing as well as
future connections created on the endpoint.
An example of this
convention could be:
where "R" is the name of the package and "qa" is the name of the
When processing a command using the "all connections" wildcard, the
"*" wildcard character applies to all current and future connections
on the endpoint, however it will not be expanded.
If a subsequent
command either explicitly (e.g., by auditing) or implicitly (e.g., by
persistence) refers to such an event, the "*" value will be used.
However, when the event is actually observed, that particular
occurrence of the event will include the name of the specific
connection it occurred on.
The wildcard character "$" can be used to denote "the current
connection".
It can only be used by the Call Agent, when the event
notification request is "encapsulated" within a connection creation
or modification command.
When this convention is used, the gateway
will generate or detect the event on the connection that is currently
being created or modified.
An example of this convention is:
When processing a command using the "current connection" wildcard,
the "$" wildcard character will be expanded to the value of the
current connection.
If a subsequent command either explicitly (e.g.,
by auditing) or implicitly (e.g., by persistence) refers to such an
event, the expanded value will be used.
In other words, the "current
connection" wildcard is expanded once, which is at the initial
processing of the command in which it was explicitly included.
The connection id, or a wildcard replacement, can be used in
conjunction with the "all packages" and "all events" conventions. For
example, the notation:
can be used to designate all events on all current and future
connections on the endpoint.
However, as mentioned before, the use
of the "all packages" and "all events" wildcards are strongly
discouraged.
Signals are divided into different types depending on their behavior:
* On/off (OO):
Once applied, these signals last until they are
turned off.
This can only happen as the result of a reboot/restart
or a new SignalRequests where the signal is explicitly turned off
(see later).
Signals of type OO are defined to be idempotent, thus
multiple requests to turn a given OO signal on (or off) are
perfectly valid and MUST NOT result in any errors.
signal could be a visual message-waiting indicator (VMWI).
turned on, it MUST NOT be turned off until explicitly instructed to
by the Call Agent, or as a result of an endpoint restart, i.e.,
these signals will not turn off as a result of the detection of a
requested event.
* Time-out (TO):
Once applied, these signals last until they are
either cancelled (by the occurrence of an event or by not being
included in a subsequent (possibly empty) list of signals), or a
signal-specific period of time has elapsed.
A TO signal that times
out will generate an "operation complete" event.
A TO signal could
be "ringback" timing out after 180 seconds.
If an event occurs
prior to the 180 seconds, the signal will, by default, be stopped
(the "Keep signals active" action - see Section 2.3.3 - will
override this behavior).
If the signal is not stopped, the signal
will time out, stop and generate an "operation complete" event,
about which the Call Agent may or may not have requested to be
If the Call Agent has asked for the "operation complete"
event to be notified, the "operation complete" event sent to the
Call Agent SHALL include the name(s) of the signal(s) that timed
out (note that if parameters were passed to the signal, the
parameters will not be reported).
If the signal was generated on a
connection, the name of the connection SHALL be included as
described above.
Time-out signals have a default time-out value
defined for them, which MAY be altered by the provisioning process.
Also, the time-out period may be provided as a parameter to the
signal (see Section 3.2.2.4).
A value of zero indicates that the
time-out period is infinite.
A TO signal that fails after being
started, but before having generated an "operation complete" event
will generate an "operation failure" event which will include the
name of the signal that failed.
Deletion of a connection with an
active TO signal will result in such a failure.
* Brief (BR):
The duration of these signals is normally so short
that they stop on their own.
If a signal stopping event occurs, or
a new SignalRequests is applied, a currently active BR signal will
However, any pending BR signals not yet applied MUST be
cancelled (a BR signal becomes pending if a NotificationRequest
includes a BR signal, and there is already an active BR signal). As
an example, a brief tone could be a DTMF digit. If the DTMF digit
"1" is currently being played, and a signal stopping event occurs,
the "1" would play to completion.
If a request to play DTMF digit
"2" arrives before DTMF digit "1" finishes playing, DTMF digit "2"
would become pending.
Signal(s) generated on a connection MUST include the name of that
connection.
2.2 Usage of SDP
The Call Agent uses the MGCP to provide the endpoint with the
description of connection parameters such as IP addresses, UDP port
and RTP profiles.
These descriptions will follow the conventions
delineated in the Session Description Protocol which is now an IETF
proposed standard, documented in .
2.3 Gateway Control Commands
2.3.1 Overview of Commands
This section describes the commands of the MGCP.
The service
consists of connection handling and endpoint handling commands.
There are currently nine commands in the protocol:
* The Call Agent can issue an EndpointConfiguration command to a
gateway, instructing the gateway about the coding characteristics
expected by the "line-side" of the endpoint.
* The Call Agent can issue a NotificationRequest command to a
gateway, instructing the gateway to watch for specific events such
as hook actions or DTMF tones on a specified endpoint.
* The gateway will then use the Notify command to inform the Call
Agent when the requested events occur.
* The Call Agent can use the CreateConnection command to create a
connection that terminates in an "endpoint" inside the gateway.
* The Call Agent can use the ModifyConnection command to change the
parameters associated with a previously established connection.
* The Call Agent can use the DeleteConnection command to delete an
existing connection.
The DeleteConnection command may also be used
by a gateway to indicate that a connection can no longer be
sustained.
* The Call Agent can use the AuditEndpoint and AuditConnection
commands to audit the status of an "endpoint" and any connections
associated with it.
Network management beyond the capabilities
provided by these commands is generally desirable.
capabilities are expected to be supported by the use of the Simple
Network Management Protocol (SNMP) and definition of a MIB which is
outside the scope of this specification.
* The Gateway can use the RestartInProgress command to notify the
Call Agent that a group of endpoints managed by the gateway is
being taken out-of-service or is being placed back in-service.
These services allow a controller (normally, the Call Agent) to
instruct a gateway on the creation of connections that terminate in
an "endpoint" attached to the gateway, and to be informed about
events occurring at the endpoint.
An endpoint may be for example:
* A specific trunk circuit, within a trunk group terminating in a
* A specific announcement handled by an announcement server.
Connections are logically grouped into "calls" (the concept of a
"call" has however little semantic meaning in MGCP itself).
connections, that may or may not belong to the same call, can
terminate in the same endpoint.
Each connection is qualified by a
"mode" parameter, which can be set to "send only" (sendonly),
"receive only" (recvonly), "send/receive" (sendrecv), "conference"
(confrnce), "inactive" (inactive), "loopback", "continuity test"
(conttest), "network loop back" (netwloop) or "network continuity
test" (netwtest).
Media generated by the endpoint is sent on connections whose mode is
either "send only", "send/receive", or "conference", unless the
endpoint has a connection in "loopback" or "continuity test" mode.
However, media generated by applying a signal to a connection is
always sent on the connection, regardless of the mode.
The handling of the media streams received on connections is
determined by the mode parameters:
* Media streams received through connections in "receive",
"conference" or "send/receive" mode are mixed and sent to the
endpoint, unless the endpoint has another connection in "loopback"
or "continuity test" mode.
* Media streams originating from the endpoint are transmitted over
all the connections whose mode is "send", "conference" or
"send/receive", unless the endpoint has another connection in
"loopback" or "continuity test" mode.
* In addition to being sent to the endpoint, a media stream rec}