SIP: a signal success: is SIP the winner in the battle of the real-time session signalling protocols? The SIP Forum, as you would expect, certainly thinks so

Telecommunications International, Dec, 2003 by Trefor Davies

In the 1990s, a debate raged over which protocol would become the standard for real-time session signalling over packetbased networks. Would it be H.323, SIP (Session Initiation Protocol) or MGCP (Media Gateway Control Protocol aka Megaco)?

Each of these protocols had different origins and different supporters with differing priorities. MGCP (and subsequently MEGACO/H248), for example, came from the telco world and was designed for inter-domain control of media gateways and softswitches (ergo Media Gateway Control Processor), which is its primary use today. Further, there was a time when H.323 and then MGCP were the only options considered for the provision of voice services over the internet. Now, however, SIP is clearly protocol of choice. Internet telephony service providers traditionally based on H.323 have begun the migration to SIP, and hosted service vendors with MGCP solutions are now predicting that their traffic will be predominantly SIP within the next 12 months.

H.323: background and basics

To understand why the migration to SIP is taking place, it's worth looking at the background to H.323 and how it works on a basic level. The main points are:

* H.323 was developed in the enterprise LAN community as a video-conferencing standard and has much in common with ISDN signalling protocols such as Q.931.

* H.323 is part of a family of real-time communication protocols developed under the auspices of the ITU, the H.32x family. Each protocol in the family addresses a different underlying network architecture.

* H.323 is not an individual protocol, but rather a complete, vertically-integrated suite of protocols that defines every component of a conferencing network (such as terminals, gateways, gatekeepers, MCUs and other feature servers). The protocols that H.323 uses inclue: Q.931 (call setup); H.225 (call signaling); H.245 (exchanging terminal capabilities and creation of media channels); RAS (registration and admission control); RTP/RTCP (sequencing audio and video packets); G.711/712 (codec specification); and T.120 (data conferencing).

All these protocols must be negotiated to set up a simple point-to-point voice call, a process that involves dozens of back-and-forth messages. As a result, the H.323 architecture includes the following elements:

* Gateways to link LAN-based H.323 end-points and PSTN end-points, as well as other networks. These translate protocols, convert media formats and transfer information. These are common to both SIP and H.323.

* Gatekeepers to translate addresses, allocate LAN bandwidth and provide other control and management functions. Gatekeepers are the brains of an H.323 network and act much like SIP servers.

* Multipoint control units, which mix and distribute conference media streams for three or more H.323 terminals.

The gatekeepers, gateways and MCUs are logically separate components of the H.323 architecture but can be implemented as a single physical device.

H.323 uses a number of protocols to set up a call. In the first instance a supported client queries an H.323 gatekeeper for the address of a new user. The gatekeeper retrieves the address and forwards it to the client, which then, using H225, establishes a session with the new client. Once the session is established, another H.323 protocol, H.245, negotiates the available features of each client. Because H.323 must first establish a session before it negotiates the features and functions of that session, call setup can take a long time. The amount of delay will depend upon the type of network.

A large part of the global signalling infrastructure used for voice over IP is H.323 based and there is no question that it has served a purpose in getting the technology adopted.

However, H.323 is fundamentally an ITU standard that was forced upon the internet; SIP, by contrast, comes from internet community, namely the IETF (Internet Engineering Task Force).

SIP: background and basics

With origins in academic in the 1990s, SIP became a blip on the radar screen in 1996 when it was adopted by an IETF working group named MMUSIC (Multiparty Multimedia Session Control). As the name of the working group suggests, one of the objectives was to develop a protocol that could initiate multiparty, multimedia conference sessions. Initially its use was perceived to be in the broadcast of multicast sessions such as space shuttle launches over the M-Bone, the internet multimedia channel.

Because of its simplicity, power and extensibility, SIP was rapidly adopted for other uses across the IETF, most notably as a VoIP standard. Communications service providers had already seen H.323-based VoIP as a way to merge disparate voice and data networks. SIP now offered them a new degree of scalability, interoperability and new service perspective that was lacking in its predecessor. The interest in using SIP for VoIP began to increase in 1998 when ideas for large-scale VoIP networks started to grow.

Because the emphasis was on supporting multiparty conferences, SIP contains support for "IP Multicast" and "request forking". These are features that have no counterparts in today's telephony world and which, together with other features such as "presence", provide for a paradigm change in the ways people communicate.