Distributed Architecture and Media Gateway Control

The foregoing discussions regarding SIP and H.323 have focused primarily
on the signaling needed to establish media streams between session participants.
Although not clearly stated, it is implied that the entities generating
the signaling are the same entities that will generate the actual
media streams. In other words, we have not described a clear separation of
media from call control.
If one looks carefully at the description of SDP, however, one sees that it
is possible to indicate different addresses for the entity that sends a session
description and the entity that actually terminates the media stream. This
indicates that the separation of media from call control is possible. Moreover,
we have seen from the architectures of 3GPP Release 4 and 3GPP
Release 5 that the separation of media and call control is not only possible,
but is often desirable.
If we physically separate a call control entity from an entity that handles
media streams (such as a gateway that performs voice coding), then we
need a protocol between those two types of entities so that the call control
entity can manage the media entity for the setup and tear-down of calls.
Provided we have such a protocol, then there is no reason why one call control
device could not manage multiple media-handling devices. It would
simply be a question of the processing power of the call control device. In
such a scenario, we can envisage an architecture such as that shown in Figure
8-11, where a single controller manages multiple media-handling
devices such as media gateways (MGs). In some quarters, this separation
between call control and media is known as the softswitch architecture.

The advantages of such an approach are that MGs can be placed as close
as possible to the source or sink of the media stream, which can be of great
significance if voice is being carried on one side of the gateway at 64 Kbps
while it is carried at a much lower bandwidth over the IP network. Though
we may place the MGs close to the edge of the network, we can centralize
the call control and network intelligence. Depending on the processing
power of the controller, the required size of the various gateways, and the
cost of each type of node, it is possible to design a network that is very costefficient
both from a capital cost and operating cost perspective. Of course,
the critical requirement is that there be a fast, robust, and scalable control
protocol between the controllers and the media devices.
As it happens, several such protocols exist. The control protocol most
widely deployed in VoIP networks today is the Media Gateway Control Protocol
(MGCP), which was developed within the Internet Engineering Task
Force (IETF). This protocol, however, has been superseded by a protocol
known as MEGACO/H.248, which was jointly developed by the IETF and
the ITU. In fact, it is known as MEGACO in the IETF community and as
H.248 within the ITU; the terms MEGACO and H.248 are interchangeable.
MEGACO has been endorsed by 3GPP as the protocol of choice for gateway
control in 3GPP Release 4 and 3GPP Release 5.