But what this year’s networking collapse has taught is the continued resilience of an installed base. Time-division multiplexed services, requiring nailed-up circuits between the transmitter and receiver of communications, have remained a crucial part of global broadband networks, and will continue to play an important role throughout at least the decade.

In practice, this implies not just the continued use of "archaic" services, like leased T1 lines, in the last mile to the user. It also means that Sonet equipment in the metro backbone, supposedly en route to obsolescence, will remain popular as broadband networks move forward. It means that enterprise-level PBXes will add some support for packet switching, but will continue to support circuit services indefinitely. And it means that any access system supporting voice-over-Internet Protocol (VoIP) that fails to include alternative gateway services for analog voice may have a limited market for the foreseeable future.

One cannot help drawing analogies between the unified world promised by the IP advocates in the waning days of the 20th century and the single-protocol nirvana promised by the proponents of asynchronous transfer mode switching 10 years ago. IP supporters will howl that ATM never had a true installed base, and that the backers of 53-byte cells were trying to create an infrastructure on the basis of scientific studies.

By contrast, TCP/IP was a protocol that arose due to the popularity of Internet access. It is often married to the framing protocol of Ethernet, a 20-year-old technology that became a "people’s favorite" in the local-area network and graduated to wide-area network duties late in the 1990s. Surely, the single-IP supporters will insist, the shift of all communication traffic to IP and Ethernet represents the victory of vox populi.

Tell that to the developers of VoIP gateways, who have done their best to convince cable and DSL modem developers to offer phone service using packetized-voice phones. Such a service has been promised again and again in the last five years, only to be hit by customer push back, the result of end-user perception that "VoIP is junk."

"The Netheads have told us for years that the resiliency and dependability of circuit-oriented services represent some vestige of old-world thinking," said a developer at Qwest, who participated in U.S. West’s interactive-service efforts and who asked not to be identified. "But at the end of the day, our enterprise customers tell us not to send out a connectionless service like IP to do a connection-oriented job. Will that change over time? Yes, but the time horizon is getting longer."

The most immediate crunch is coming in the notorious "last mile" between home or business and the central office, an area we examine in the related article on wired services. Just as this access point was a bottleneck in terms of gaining true broadband services, it is also a bottleneck in how IP traffic is added to a mix previously dominated by analog circuit-switched traffic (or, in the case of cable TV, analog broadcast traffic). Telephone central offices and cable TV headends function very differently depending on whether IP traffic arrives in an aggregation with circuit-switched traffic, or via an independent overlay network. This is not just a problem for older "transition" networks of the 1990s, but will remain an architectural issue for the next 10 years.

For example, said Chris Lawler, vice president and general manager of IP routing at Unisphere Networks Inc., many people assume that when a corporation decides to move its voice traffic to IP, it places that traffic on the same general-purpose routers that carry the data traffic. This is not the case with recent deployments, he said. Many companies are building dedicated voice gateway and edge-router subnetworks for their voice traffic, and keeping these networks completely separate from data routers, until the traffic is eventually aggregated at a central office.

"The thinking is that a packet network should be created that is totally dedicated to packetized voice," Lawler said. "So when people talk about ‘convergence,’ it is not a complete convergence of IP traffic by any means."

A vestige of the old world that may remain with us for an indefinite number of years is the Class 5 electronic circuit switch. Lucent, Nortel, Alcatel, Cisco and several carriers were active in defining "soft-switch" architectures based on the Media Gateway Control Protocol and the Session Initiation Protocol. In theory, large carriers could purchase mediation switches that talked both IP and Signaling System 7, and consign those old 5ESS switches to the trash heap. In practice, it is the soft switch that almost seems headed for the dustbin at this point. Its integration into modern networks has been spotty at best.

Unisphere’s Lawler pointed out that many "mediation switch" vendors have changed business plans to emphasize the replacement of the Class 4 tandem switch in the network backbone, where it is easier to move to all IP. The assumption is that many Class 5 central-office switches will remain in place, he said.

The continued popularity of Sonet network elements in metro and supermetro regions can be seen by the fate of Cerent Corp. Two years ago, few could believe that Cisco Systems Inc. paid $6.9 billion for Cerent, while simultaneously buying optical-routing company Monterey Networks for $500 million. At the time, Monterey was considered the wave of the future, and Cerent was dismissed as a TDM relic. But Cerent platforms have proven to be a primary revenue stream for Cisco, even during the 2001 downturn. By contrast, Monterey’s prospects were pushed out so far that product plans got put on ice this past spring.

Nevertheless, the desire to simplify Sonet-layer protection switching, and limit the wasted bandwidth encountered when a TDM fiber ring like Sonet carries packet traffic, has spurred a new effort to create redundant fiber rings without the full dual-fiber redundancy of Sonet. Many companies in both Ethernet and Sonet camps have joined a new IEEE working group, 802.17, which seeks to define a resilient packet ring for creation of the most efficient metropolitan infrastructure possible.

The working group and a related industry consortium, RPR Alliance, already are butting heads with the EFM and MEF Ethernet groups, however, because 802.17 proposes a totally new media-access control layer for data-link chips, something many companies fear that carriers and end users will be reluctant to accept.

Companies on the bleeding edge of optical transport would like to bypass the RPR/Ethernet battles in the metro region by implementing direct IP transport over wavelengths. Two players, Ciena Corp. and Tellium Inc., have developed unique hybrid optical/electronic switches that perform optimal packet switching over wavelength,doing away with the need to wait for an all-optical infrastructure in the network core.

Other newcomers, such as Atoga Systems and Network Photonics, are betting on the imminent availability of tunable lasers and filters, which will allow more complex multiplexing and switching of wavelengths with a minimum of optoelectronic conversions.

Recession or no, this shift to a packet-centric optical transport has changed the market dynamics from those of the Sonet world, where vendors like Fujitsu and Alcatel dominated metropolitan transport. The Aberdeen Group released a study in early September pointing out that Ciena’s lead in hybrid switching had given the company an impressive leading role in both metro and long-haul transport. Meanwhile, the super-regional metro space was being dominated by newer players like Tellium and ONI Systems, as well as the Cerent group at Cisco.

Transcontinental and global long-haul networks are the realms where packet-over-optical may dominate first. Certainly, many of the undersea wave-division multiplexed fibers have been developed with Sonet in mind, due to the high number of TDM circuit-switched calls these networks must carry. But the bulk of new players in continental markets, such as Level 3 Communications Inc. and Global Crossing Inc., which offer a variety of wholesale transport services optimized for packets, will ensure that many of these networks are early adopters of packet-over-wavelength and generalized multiprotocol label-switching (G-MPLS) technologies.

In theory, most of the so-called excess bandwidth could be used if more end customers had access to broadband last-mile services. And the biggest overbuild of dark-fiber networks in the 1999-2000 period tended to be in metro and supermetro regions. But interexchange carriers (IXCs) do indeed face a problem: Internet service providers and application service providers are disappearing as a result of the downturn; which means that some alternative carriers are disappearing; which means that fewer wholesale customers are available for the Level 3s of the world.

The result is packet price bombing that calls into question the viability of even the best-managed of the long-haul players. Level 3 chief executive James Crowe has had to deal with an investment community that wants to devalue IXC assets across the board, including those of his company.

Crowe complained to the press in August that fiber-based IXCs had shifted in the space of a year from "poster children . . . to pietàs." While Level 3 has $3.3 billion in the bank and a global network built out in North America, Europe and Asia, the company’s ability to add next-generation equipment for soft switching is significantly hindered by Wall Street’s gloom.

The wildest card of all may be wireless services. The digital 800-MHz and 1.8-GHz networks now handling the bulk of cellular calls retain a circuit-switched infrastructure, and the packet-based services promised for the 2.5 generation do not change that analog baseline. The turn to packetized voice and third-generation frequency bands was supposed to take place in tandem. The overspending of carriers on 3G licenses has meshed with the lack of understanding of IP services among those same carriers to postpone indefinitely the turn to digital packet technology.

To be sure, some infrastructure equipment providers serving the wireless area, particularly Nortel, are making the case for moving to a packet-based switching system now, and can make the economic case for carriers that are looking at new regional buildouts.

The problem comes in regions with existing services. Carriers stretched to the breaking point in recouping 3G auction investments are being asked to spend on new antenna infrastructures, new basestation switching centers, new home and visitor location registers and new antenna-planning software.

If wireless carriers are familiar with VoIP gateways, they can move to some IP switching backbones now, without worrying about how long it takes to shift to 3G. And of course, the fixed-wireless service providers giving an option of broadband wireless connections to the home or small office are familiar with IP from the beginning, since the bulk of all metro and local multipoint distribution service traffic is IP.

But shifting the bulk of cellular voice traffic to native IP is gated by the acceptance of 3G packet services. As our writers point out in this issue, it will be a long and winding road to wireless IP.