Softwitching Seeks Seeks IP, PSTN Fusion.

Protocols could be its Achilles Heel 

The use of standards-based software to control interoperable hardware has always been a key design goal of network architects. Updating proprietary software and hardware is difficult and expensive, and it's time consuming. In the telephone network, where routing is primarily handled by digital circuit switches, updating the underlying architecture is a particularly expensive and complex process. That's why networking companies, carriers, and telco equipment suppliers are seeking to simplify the overall design of the telephone network with softswitches. The hope is that softswitches will enable easier connection of the phone network to IP, provide cheaper, faster upgrades, and spur application development.

Softswitching technology follows a basic underlying premise: that by using IP-based protocols and standard Java or C++ development environments, the PSTN will be easier to upgrade with new data services as they are introduced. The companies behind the softswitch movement are seeking to separate telephone switching into non-proprietary structural elements: standard software on one side, and interoperable--that is, non vendor-specific--hardware on the other. This concept is the bedrock on which computer networking technology is based.

In the Public Switched Telephone Network, interoperability is uncommon. For example: The majority of telephone calls are handled by Class 4 and 5 circuit switches, which handle call routing, status, and distribution. Class 4 and 5 switches traditionally use proprietary hardware and software, and even proprietary development environments. For new services (and hence new routing functions) to be added, significant upgrades are necessary, and the rule of thumb is an 18 month turnaround time for new services to be rolled out.

Developers creating applications for new services are constrained by 1) the need for expensive switch upgrades; and 2) the lack of interoperability among switching software. Obviously, a delay of 18 months for new services is unacceptable in an IP-based network like the Internet. But is it necessarily so in the PSTN? Further, can softswitches provide the full feature set--between 270 and 350 call routing functions--of Class 5 switches?

Softswitching technology--which, not incidentally is being pushed by its own consortium; more on this later--is offering two "solutions" to circuit switching. One is a series of standard protocols and a standardized development environment that vendors and developers can use in building hardware and writing applications. The other is the development of the actual softswitches themselves, which operate by distributing the functions currently performed by digital circuit switches.

Switching Asides

The primary goal of the softswitching movement is to seamlessly connect the PSTN and IP. Of course, in order to do this, protocols need to be adapted and adopted. While some are already established, others are being tested by the Softswitch Consortium, a group of dozens of companies that includes founders Cisco, Lucent, Telcordia, Level 3, Nortel, and HP, among others. The group advocates interoperability and certification based on H.323 protocols. However, the Softswitch Consortium also promotes interoperability based on the Session Initiation Protocol (SIP), the Media Gateway Control Protocol (MGCP), and Real-time Transport Protocol (RTP). According to the Consortium, the group intends to develop new standardized interfaces for portable applications which will sit on top of any based softswitched network.

In particular, SIP seems to have significant momentum. SIP is a text-based protocol, similar to HTTP and SMTP, for initiating interactive communication sessions between users, according to Henning Schulzrinne. Schulzrinne is head of the Internet Real-time Lab and an Associate Professor in the departments of Computer Science and Electrical Engineering at Columbia. SIP-initiated sessions might include voice, video, chat, interactive games, and virtual reality. SIP is currently an IETF proposed standard, RFC 2543; a draft standard is expected by the middle of 2000. Predictably, products based on SIP and SIP+ (a version with feature extensions) have already been announced.

SIP is independent of the packet layer and only requires an unreliable datagram service; it provides its own reliability mechanism. According to Schulzrinne, while SIP typically is used over UDP or TCP, it can, without technical changes, be run over IPX, frame relay, and ATM, in rough order of desirability. (This order has significant implications for the PSTN, as discussed below.) SIP was first developed within the Multiparty Multimedia Session Control (MMUSIC) working group of the IETF. It is a flexible and powerful protocol that can be used in a number of ways for managing calls. It can be used to send session description data, to set up calls, and to talk to other softswitches on the network.

Softswitching takes much of the routing burden off the switch itself. "With softswitches, services can be distributed throughout the network," says Mike Clement, senior product manager at Unisphere Solutions, a maker of softswitches. "They can even be moved from the switch all the way to the edge of the network, to the gateway. Call management functions can be handled by software written with standard programming methods and languages such as C++." These functions are then transported by a set of protocols that support IP. The Fig shows the various protocols and services associated with softswitching.

It's The Protocol, Stupid

Softswitch vendors like to point to the benefits of switches based on standard software: interoperability, faster addition of new services, and cost savings. But some observers feel that softswitches have the same inherent weaknesses as traditional Class 4 and 5 switches: their reliance on protocols.

"The problems associated with updating services in the PSTN will not disappear with softswitches," says analyst Tom Nolle. Nolle is president of CIMI Corp., a networking consultancy with clients that include Lucent, Nortel, and Cisco. "The problem is in the underlying protocols. Existing PSTN switches use SS7 as the transport protocol. IP simply replaces this with SIP. The issue is that when new services are added, it's the protocols that have to be updated to support them. SIP and softswitches do nothing to change this."

But the problem with softswitching, Nolle feels, is in the technology itself, not in how it is implemented. The issue is that the PSTN has always been, and continues to be, a state-based network: switches must keep track of the state of a call at all times. For example, when a second call (known as a "flash") comes into a phone line with call waiting service, the switch routing the first call must keep track of it, making sure that the initial call is not terminated when the user picks up the incoming call.

Since IP is a "stateless" protocol, additional protocols like RSVP and RTP must be used to keep track of call states for services, and especially QoS-based applications that demand specific bandwidth. And while call state can be distributed in a softswitched net-work--as opposed to being maintained solely at the switch level--distribution can be a complex procedure.

"Distributed call state is more powerful because more services can be offered to the caller," says Eric Zimmerer, director of product management at ip Verse Inc. and chairman of the Softswitch Consortium's SIP Working Group. "It is also more difficult to implement because of the problems associated with maintaining call state in multiple places. What do you do if one box in the system fails? How do you know if you are in sync with every network element that has touched the call? A lot of bandwidth and CPU cycles can be wasted comparing and double-checking distributed call state tables."

Zimmerer notes that, depending on the circumstances and the application, there are clear advantages to having call state maintained by the switch. "There is the issue of intelligence at the endpoint vs. intelligence at the center of the network," he says. "Intelligent endpoints such as SIP-[based] phones do not require any call state to be kept in the network because each [phone] keeps track of the call state. This turns the network into a 'dumb pipe' that merely passes bits back and forth."

While he notes that this scenario may be very attractive to the technically adept user, Zimmerer acknowledges that the vast majority of users derive benefit from an intelligent network where the end-points are "dumb," such as today's PSTN and our POTS phones. "Intelligent networks keep track of call state in the process of providing (and billing for) services," he says. Most end users don't want to worry about the complexity behind the network; they just want the phone to work when it's picked up.

All of this complexity is being added and then used to shoehorn IP into the PSTN at a time when, ironically, all the major RBOC carriers are moving their trunks to ATM anyway. "The entire telephone network, from the phone to the CO [Central Office, where Class 5 switches are located] is going to ATM, and no one is going to care about things like VoIP with ATM, because ATM doesn't need IP to carry voice," Tom Nolle states. Further, ATM already supports IP and provides services like state monitoring and QoS provisioning, making many of the protocols behind softswitching redundant. ("It's like routing your phone call over your dial-up modem," Nolle jibes.)

Is there evidence to support Nolle's assertion? Actually, there is. In November, Lucent announced that SBC (the largest of the RBOCs) would purchase $2.4 billion worth of Lucent Class 5 switches. The interface on the switches? ATM. The purchase is part of SBC's Project Pronto, which the company claims will make broadband services available to 80 percent of its customers over the next three years. Project Pronto will "accelerate the convergence of [SBC's] voice and data backbone systems into a next-generation, packet-switched platform." SBC intends to deploy "the industry's most advanced packet-switching technology: voice trunking over Asynchronous Transfer Mode." Project Pronto is the largest single ATM implementation in the world to date.

In a separate announcement, KMC Telecom indicated that, in a deal valued at nearly $1 billion, the company will buy Lucent hardware to expand its voice and data network in small and medium-size cities. The deal will enable KMC to offer VoIP service in cities where Lucent's 5ESS (Class 5) digital switches are already in use. The upgrade to Lucent's PathStar Access Server and Stinger, a new DSL access concentrator, will give KMC customers voice and data services over DSL using--you guessed it--an ATM backbone. Lucent's PathStar system integrates the local loop termination, a Class 5 switched telephony system, a VoIP gateway, and an edge router into a single network element.

What does this migration to ATM mean as far as call routing goes? For one thing, it means that all packets, be they voice or data, will hit an ATM trunk and be transported via ATM to their end-office switch. It probably also means that separate virtual ATM circuits will be provisioned for voice and for data packets, at least for the foreseeable future. For consumers, the only difference will be that they will have broadband voice and data services coming into (and going out of) one pipe--probably a DSL card in the near term.

But if softswitching and VoIP technologies are being subsumed by ATM at the carrier level, why then are so many major networking vendors (Cisco comes to mind) and countless startups behind them? One reason may simply be that these companies are afraid of what their businesses might look like if IP doesn't turn out to be the PSTN killer many hoped it would be. A second is that there are enormous amounts of money to be made--potentially billions--in selling softswitches to telcos and telecomm carriers. But first, the telcos need to be sold on softswitching. And, at present, the PSTN certainly ain't broke; many are asking what exactly softswitching hopes to fix. "There is a quarter-trillion [dollars] invested in the domestic telephone infrastructure," says Tom Nolle, "The Internet community's vision of the Internet as a public phone network is just bogus."