Inside the Session Initiation Protocol

Network managers getting ready to roll out IP telephony networks had better think twice about the gear they're going to deploy. A new IETF protocol may change the way next -generation phone networks are built. The Session Initiation Protocol (SIP) could provide the underlying mechanisms for establishing calls between users on an IP telephony network.

That might not seem like a big deal. After all, the H.323 suite of protocols has long provided comparable functionality. However, SIP's smaller footprint makes the protocol more scalable and faster than existing H.323 implementations . The catch? The protocol is still in its early stages, making products hard to come by.

Until recently, network managers looking to roll out intelligent networks have relied heavily on the H.323 suite of protocols. With H.323, a compliant client, such as Microsoft NetMeeting, 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 establishes a session with the new client using H.225, one of the H.323 protocols. Once the session is established, another H.323 protocol, H.245, negotiates the available features of each client.

It may sound simple, but H.323 suffers from some key problems. At the top of the list is call setup time. Since H.323 first establishes a session and only then negotiates the features and capabilities of that session, call setup can take significantly longer than an average PSTN call (see Figure 1).

Figure 1: The time needed to set up a call is the key problem with H.323. Since the features of the call, such as whether or not to invoke video, are only negotiated after the call is established, setup times are much longer with H.323 than with SIP.

Just how long depends on the particular network and the distance between locations, but the total time for someone to answer the call can reach up to 8 seconds, according to Pauli Saksanen, project manager for IP telephony at Sonera (www.sonera.com), the former Postal, Telegraph & Telephone (PTT) administration of Finland. The delays are even worse on international calls, he says, with lag times as long as several seconds. Saksanen should know. He runs Sonera's nationwide IP telephony network, the first of its kind anywhere in the world.

The network is currently based around H.323, but Saksanen says it will ultimately be able to accept clients using other protocols as well. By working with existing component suppliers, Saksanen says he's been able to get the delay down to between 100ms to 200ms, but the problem still exists. A new version of H.323, dubbed "H.323 fast," will address the problem, but H.323 fast isn't widely accepted yet.

What's more, opponents say H.323 doesn't scale well. A case in point is H.323 addressing. Creating separate phone-numbering schemes complicates interconnecting carrier networks. Critics also charge that the H.323 standard itself is too large and complex to make deployment easy. "H.323 is built in a telecom manner," says Hans Eriksson, chief technology officer at Telia Network Services (www.telia.com), a division within Telia, the Swedish PTT. Eriksson has evaluated H.323 as a way of rolling out telephony services over an IP network. Finally, H.323 doesn't provide a simple way for connecting two circuit-switched networks across an IP network.

Enter SIP

All of these problems are addressed by SIP. With SIP, each user is identified through a hierarchical URL that's built around elements such as a user's phone number or host name (for example, SIP:[e-mail]). The similarity to an e-mail address makes SIP URLs easy to guess from a user's e-mail address.

When a user wants to call another user, the caller initiates the call with an invite request. The request contains enough information for the called party to join the session. With a unicast session, this includes the media types and formats that the caller wants to use and a destination for the media data (see Figure 2). A session might include, for example, sender requests to employ H.261 video and G.711 audio.

Figure 2: Session Initiation Protocol (SIP) speeds up call setup by bundling all of the configuration information in the request. The request is either sent directly or via the proxy server to the recipient, who accepts the call and immediately initiates the session.

The request is sent to the user's SIP server. The SIP server may be a proxy server, which receives the request and, using its own internal algorithms, determines the user's location. Alternatively, the SIP server may be a redirect server that returns to the client the appropriate SIP URL, which the user then queries. In either case, the server's address is learned by querying the DNS, the distributed database that matches high-level host names with the underlying IP address.

Once found, the request is sent to the user, and from there several options arise. In the simplest case, the request is received by the user's telephony clientthat is, the user's phone rings. If the user takes the call, the client responds to the invitation with the designated capabilities of the client software, and a connection is established. If the user declines the call, the session can be redirected to a voice mail server or to another user.

"Designated capabilities" refer to the functions that the user wants to invoke. The client software might support videoconferencing, for example, but the user may only want to use audio conferencing. Regardless, the user can always add functionssuch as videoconferencing, whiteboarding, or a third userby issuing another invite request to other users on the link.

SIP has two additional significant features. The first is SIP's ability to split, or "fork," an incoming call so that several extensions can be rung at once. The first extension to answer takes the call. This feature is handy if a user might be working between two locations (a lab and an office, for example), or where someone is ringing both a secretary and a boss.

"You could extend H.323 to offer that feature, but under the existing standard, forking just isn't possible," says Henning Schulzrinne, associate professor in the departments of computer science and electrical engineering at Columbia University in New York, and one of the original authors of the SIP standard.

The second significant feature is SIP's unique ability to return different media types. Take the example of a user contacting a company. When the SIP server receives the client's connection request, it can return to the customer's phone client via a Web Interactive Voice Response (IVR) page, with the extensions of the available departments or users provided on the list. Clicking the appropriate link sends an invitation to that user to set up a call.

Border Patrol

With the basics covered, it's easy to see where SIP fills in some of H.323's holes. First there is the issue of call setup time. By including a client's available features within the invite request, SIP negotiates the features and capabilities of the call within a single transaction. The upshot, says Schulzrinne, is that SIP can set up a call within about 100ms, depending on the network.

SIP also scales better than H.323. One of the attractive features for providers like Telia is SIP's simplicity. It doesn't purport to solve the whole telephony equation, says Eriksson. He notes SIP can be used just to identify an end user, relying on other protocols and applications to manage the call. The result, says Eriksson, is that the protocol is much easier to implement than H.323 and much easier to adapt.

A case in point is SIP's addressing scheme, which leverages the existing DNS system instead of recreating a separate hierarchy of telephony name servers. Then there's the way SIP handles connectivity with circuit-switched networks. A new SIP draft will also extend the protocol to address a number of other problems. Chief among them is the need to develop a standard way to map the telephony service parameters onto SIP packets using the MIME standard. By using MIME for signaling, the data ends up being passed along on a SIP transaction in much the same way that e-mail attachments are transported with a mail message.

The protocol's light weight lets it be enhanced in other ways as well. The draft standard is expected to add several special-purpose functions to SIP. Among these options is an ability to negotiate security and QoS. There's also the ability to let callers indicate their various preferences. A user calling a company, for example, might want to speak only with someone who speaks Spanish. This option can be embedded in the Invite command so that the user will be routed to the correct company contact.

Are there drawbacks to SIP? No question. The biggest issue today is availability. While H.323 is widely accepted and deployed today, SIP products are, for the most part, nowhere to be found. That will soon change, however. Currently, as many as 20 vendors are working on SIP implementations. 3Com announced last October that it's including SIP support within its CommWorks IP telephony solution for service providers. Cisco Systems is doing the same by including SIP support in its Architecture for Voice, Video and Integrated Data (AVVID) architecture. And while Nortel Networks has not announced SIP support directly, the company is collaborating with Telia Mobile on developing next-generation telephony services utilizing the protocol.

That's a pretty strong vote of confidence, but it may not be enough to compensate for the Microsoft factor. "Users already get an H.323 client today with every version of Windows," says Sonera's Saksanen. "Will they be willing to go out and purchase a SIP client as well?" That's the question.

Resources

The actual Session Initiation Protocol (SIP) standard (RFC 2543), submitted by Mark Handley, Henning Schulzrinne, Eve Schooler, and Jonathan Rosenberg, is at ftp://ftp.isi.edu/in-notes/rfc2543.txt/.

Henning Schulzrinne, one of the original authors of the SIP standard, has an excellent site at www.cs.columbia.edu/~hgs/sip/sip.html. There's an overview of SIP, a schedule of the interoperability tests (bake-offs) of SIP gear, plenty of technical details on the protocols, and links to associated sites. If you're worried about grammar (SIP grammar, that is), the site provides all the gory details on how to format SIP requests.

SIP implementers can stay current with the latest information by subscribing to the list at [e-mail]. Send mail to the address with the line "subscribe sip-implementers" in the body.

This tutorial, number 138, by David Greenfield, was originally published in the January 2000 issue of Network Magazine.