August 1, 2004

SIP Primer for Cable

By Justin J. Junkus, Telephony Editor

The session initiation protocol (SIP) has become a buzzword in cable voice over Internet protocol (VoIP) recently. Vendors have made SIP famous in applications ranging from adapters for telephony over a high-speed data interface through telephone sets to deliver multimedia in small office-home office environments. This column is a guide to SIP that should prove helpful in understanding where SIP fits into cable VoIP and PacketCable.

Defining SIP
The Internet Engineering Task Force (IETF) defined SIP in RFC 2543 as a signaling protocol. It is similar to the network control signaling (NCS) protocol that is the foundation for most of the early cable VoIP telephony products being used in telephony introductions or trials. The difference is that SIP signaling is done by endpoints. While NCS depends upon call management servers (CMSs) within the service provider’s network to set up a session such as a telephone call, SIP assumes an endpoint that has enough intelligence to handle this task on its own. Because SIP does not manage transport, its job stops with signaling, and moving the actual content of a session between endpoints is the job of other protocols such as real time protocol (RTP). In fact, content takes an entirely different path across the network than signaling.

The original SIP RFC has been updated since 1999, and several IETF workgroups have created additional SIP documents dealing with issues like security. A listing of IETF work is posed at www.ietf.org/html/charters/sip-charter.html.

Referring to the three-layer VoIP model I discussed in April, SIP signaling can be applied to access, distribution and core. Recent product noise has been in the access arena with SIP-based telephones, handhelds and even wireless devices. However, at the SIP Summit in Chicago this June, Jean-Francois Mule, director of PacketCable architecture, said cable has also applied SIP in distribution and core via the Call Management Server Signaling (CMSS) PacketCable specification. PKT-SP-CMSS-i02-021205 defines how cable communications networks use a form of SIP to communicate between themselves in setting up calls across an internet.

How SIP works
SIP architecture and signaling messages are relatively simple. A SIP system is defined by user agents and network servers. User agents are client-end systems applications that contain both a user-agent client (UAC) and a user-agent server (UAS). The UAC initiates SIP sessions and is called the user’s calling agent. The UAS receives requests for session activity and returns responses on behalf of the user. It is also called the user-called agent.

Network servers accept SIP messages from user agents and facilitate message routing. There are three types of SIP network servers: registrar servers, proxy servers and redirect servers.

Registrar servers keep track of SIP UACs. When a UAC comes on line, it sends a registrar server its SIP address, which can be a domain name (for example, sip: jjunkus@knowledgelinkinc.com), an E.164 (PSTN) phone number, (for example, 16308208205@gateway.com; user = phone), or both. The registrar server provides this address to a database called a location server, which is used to find end users that have changed endpoints. Users may register multiple addresses to conduct a session from one or more devices.

Proxy servers receive SIP signaling messages and forward them to the next SIP server to establish an end-to-end signaling connection. They also may perform authentication and authorization. Redirect servers, rather than forwarding SIP messages, look up the next hop address and send it back to the UAC or previous proxy server for use in forwarding the SIP message. Proxy and redirect servers may be combined in hardware.

Signaling is done by six types of signaling request messages and six categories of response messages. In a typical signaling path setup, a session is initiated by an INVITE request message, and upon completion of the signaling path, the called party user agent sends back an OK response message. As a further handshake in this process, the calling party returns an ACK request message, indicating it received the response. Once the call has been set up, the signaling path is torn down, so the disconnect sequence of signaling (started by a BYE message) will most likely travel a different path than the setup signaling.

Real-world applications
A good part of what makes SIP so exciting is the ease with which it melds into data applications. SIP is a text-based protocol similar to hypertext transfer protocol (HTTP), which is the basis for the World Wide Web. The call processing language (CPL) associated with signaling messages is similar to hypertext markup language (HTML), which is used to define Web page layouts. Signaling messages can contain JAVA applets, images, audio files, authorization tokens, or billing data, which creates endless possibilities for multimedia even during the setup process.

The intelligence at SIP endpoints permits call processing to be done there, rather than within the network, so functions such as routing a call to ring or voice mail based upon who is calling can be programmed at the user device. Many HTML programmers can create CPL-based services, and third-party graphical user interfaces (GUIs) developed for Web applications potentially can be reused with SIP. This makes user service modification and service provider service creation a lot closer to realization. In addition, the registration process helps a user experience become device-independent, leading to new “presence” applications that follow a user to the nearest device.

What’s the holdup?
Despite the benefits, there are several reasons SIP was not the first choice protocol for PacketCable end points. Cable needed to get into the residential VoIP phone business quickly and at a reasonable cost per sub. The intelligence needed at SIP end points implies a more complex device than a multimedia terminal adapter (MTA). NCS signaling presents a cost-effective solution for carrier-grade residential telephony, but without full multimedia functions.

As silicon costs decline and user demand for functionality increases, SIP-based endpoints become more attractive. While this is happening, some key issues remain unresolved. Emergency services (E911) and federally mandated wiretapping solutions (CALEA) are not yet standardized for SIP, and security-related issues are pending. Firewalls and network address translation (NAT) remain a problem. Because SIP is signaling only, quality of service (QoS) issues must be solved by other protocols that must interwork with SIP.

Because of these issues, the main applications of SIP in cable today are related to using our high-speed data transport as a pipe for applications delivered by other service providers. We are the transport for the Internet telephone provider who interfaces with residential customers via a SIP adapter, or the means for extending a corporate virtual private network (VPN) to SIP endpoints for work at home. The trend is for cable to move closer to revenue-generating applications, but only after we can ensure service guarantees consistent with carrier-grade solutions.

Justin J. Junkus is president of KnowledgeLink Inc. Email him at jjunkus@knowledgelinkinc.com.