Network Computingwww.networkcomputing.com

As I've written before, the industry is failing with H.323 ("When Good Standards Go Bad," www.networkcomputing.com/1017/1017colwillis.html). The ITU-T has developed a thick series of labyrinthine specs that are too expensive for anyone to implement. But don't expect H.323 to go away: Sheer momentum will keep it alive, with its best success in trunking and toll-bypass apps.

Still, H.323 won't improve the user experience anytime soon. It has failed to produce next-generation endpoints or advanced features in virtual PBXes. To support innovation, the industry must simplify development with common software interfaces and high-level tools. But even that's not enough, as CTI products show: While component manufacturers have greatly simplified development, ISVs are still cranking out proprietary systems. What's lacking is a data network-based telephony services model that can quickly provide usable, interoperable solutions.

To bring advanced voice support into the data networking age, protocols must be targeted directly at actual user services. Products should easily integrate into a real network with little modification to its underlying infrastructure. The protocols should be easily extensible without breaking existing implementations or relying on a factious standards body for approval.

Based on these criteria, the IETF's SIP (Session Initiation Protocol) looks like the winner in the voice over packet service creation race. Defined in RFC 2543, SIP establishes, modifies and terminates voice calls. It borrows from established protocols, such as HTTP and SMTP, and is easy to analyze on the wire. It is readily extended, with many proposals floating about just a few months after SIP's approval as an RFC. Lucent, Cisco, 3Com and Ericsson are in on the action, as are CableLabs, Telcordia, General Instrument and Com21. AT&T, MCI and Level 3 have been involved in SIP's development. Microsoft is conspicuously absent.

SIP's simplicity doesn't compromise its power. The authors address key VoIP features including encryption and authentication. SIP's client/server orientation offers a level of server-based call management missing in the peer call model that most H.323 endpoints use. The first thing a SIP client does is locate a server, typically via DNS. SIP proxies can be easily integrated into firewalls and Network Address Translators. Proposed SIP extensions include specs for user-based call-security management and QoS requirements, as well as the signaling of changes in network conditions.

Many standards never yield successful products, but SIP is not just academic. Many SIP implementations are on the market as single-line gateways, proxy servers, media gateways, Java toolkits and native endpoints. HP's Internet Advisor analyzer already provides SIP decodes. At least two product interoperability bake-offs will have occurred by the time you read this.

VoIP must do more than reduce toll charges. It should offer new capabilities far beyond those of the current PSTN and close the gap between what the network offers and what users want. Simplicity will be the key to fast development, and the most promising key to simplicity is SIP.

Send your comments on this column to David Willis at dwillis@nwc.com.