The fiber-optic glass may be mostly empty, but it’s filling up fast. Back in the mid-1990s-an eon in Internet time-telecom pundits predicted that the Web of the early 21st century would run on pure fiber-optic cable, delivering massive bandwidth and a raft of powerful applications right to the door of businesses. Broadcast-quality desktop videoconferencing, heavy-duty data mining, instant disaster recovery, jitter-free Internet phone calls; all would be available to even small companies once Internet Protocol (IP) packets rode the light, unencumbered by artifacts of the previous century, like copper wires and circuit switches.
The promise of the optical Net remains unrealized today. The telecom meltdown has withered investment, pushing carriers that built their own fiber networks, like 360networks, XO Communications, and Yipes Communications, into bankruptcy. A confusing patchwork of new and legacy technologies stitches fiber networks together. For over 90 percent of businesses, Internet traffic still begins and ends its journey on twisted-pair copper or coaxial lines. Nevertheless, the migration to IP-over-fiber is under way, laying the foundation for ultra-fast, streamlined optical networks destined to erase the current distinction between Internet traffic and other forms of telecommunication.
Leveraging existing infrastructure, regional telecom providers such as CTC Communications and Onvoy Inc. have engineered IP-over-fiber backbones that allow customers to transmit phone conversations, streamed video, private data, and Internet traffic over the same pipe, bypassing the traditional public switched telephone network (PSTN). “[Converged services] will change the Internet dramatically, and the way that people interface with the Internet,” says Tom Jenkins, vice president of consulting for TeleChoice, a telecom consulting firm based in Tulsa, Okla.
Hopeful that they can avoid the fate of Yipes, surviving IP-over-Ethernet carriers continue to expand their presence in select cities, offering bandwidth-hungry customers high-speed Internet access and data services piped over fiber directly into office buildings. In experiments with next-generation fiber switches and faster flavors of fiber, researchers are making progress toward “pure” IP-over-fiber technology that would deliver virtually limitless bandwidth.
But if the optical Net ever is to become a reality for the business mainstream, the Baby Bells that control the “last mile” between fiber backbones and end-users must embrace the concept of Internet over glass.
Convergence: one pipe for all
The great virtue of IP–the reason why telecom companies have chosen it to power their new fiber networks–is its flexibility; unlike circuit-switching, which was invented more than a century ago to transmit phone conversations, IP can transport any form of digitized information as a series of packets, or datagrams. Using IP, a carrier can converge its traffic, shunting voice, video, and private data along the same glass pathways traveled by e-mails and elements of Web pages.
That’s much more efficient because IP utilizes the entire bandwidth available on a fiber strand, eliminating wasted capacity. It’s cheaper because telecom providers and enterprises needn’t maintain separate systems for voice, video, and data communication. And IP-over-fiber opens the door to emerging and as-yet-unimagined applications with the potential to vastly enhance productivity: toll-free long distance; unified messaging; video whiteboarding; truly intelligent, Web-browsing phones. Even the proverbial videophone, prototyped in the 1960s, enters the realm of possibility once video and voice share the same medium.
More and more telecom companies are staking their future on convergence, betting that IT managers will turn to them for one-stop-shop convenience and bottom-line savings.
“For some providers, their only goal with launching a convergence solution is to gain efficiencies, lower their own cost point,” Jenkins says. “Others are doing it to enter new markets, to attract new customers, to offer new services to existing customers to get their average revenue per user up.” CTC Communications , an integrated communications carrier based in Waltham, Mass. with 15,000 business accounts along the Eastern Seaboard, sees convergence as a way to differentiate itself from local-loop juggernauts such as Verizon Communications and SBC Communications. “Going after them with a smaller customer base and the same tools didn’t seem to be a very good business strategy,” says Russ Oliver, CTC’s vice president of network operations.
Since 1999 the company has spent $180 million building its PowerPath Network, an IP-over-fiber Web extending from Maine to Virginia. Once customers–many of them mid- to large-sized enterprises with multiple locations–install integrated access devices on site, IP phone calls, private data, and Internet traffic come to them across a single broadband connection. Oliver says that PowerPath’s efficiency slashes customers’ telecommunication expenses by up to 40 percent, while boosting CTC’s profit margins to more than 50 percent. CTC is banking on numbers like that to help the company stem a tide of red ink ($124 million lost in 2001) and a return to profitability next year.
Similar motivations are behind Onvoy’s Converged IP Services rollout in Minnesota this fall. The suburban Minneapolis carrier, a major provider of voice, video, and data transmission services to business and government in the Upper Midwest, has parlayed its extensive fiber network and experience as an ISP into an all-IP network featuring voice-over IP (VoIP), digital videoconferencing, VPNs, and dedicated Internet access. “Frankly, bundling all of that stuff together in a cost-effective IP network was a natural play for us,” says CEO Janice Aune. Investors have backed the initiative with a $25 million infusion of capital.
Other carriers pursuing IP convergence include Sprint Corp., British conglomerate Cable & Wireless, Cbeyond Communications in Atlanta, and Dallas-based Allegiance Telecom.
Embracing IP has required telecom companies to tweak their fiber networks, moving them closer to the notion of IP wedded to pure fiber, stripped of intervening framing mechanisms and transport protocols that compromise bandwidth. “Onvoy and others doing converged services need to move IP packets, and find the best way to do that,” says Dennis Fazio, principal consultant with HeatSeeker Technology Partners Inc., a Minneapolis-based IT infrastructure consulting firm. “That may take a couple of generations of evolution, all of which should be invisible to the users of the converged services.” Carriers aren’t totally revamping their fiber plants; they’ve got too much invested in legacy equipment to throw it all out and start over. But they are migrating from older technologies such as time-division multiplexing and ATM, to more data-friendly, less costly systems that expedite the flow of IP packets over dense wave-division multiplexed (DWDM) fiber.
Onvoy, for example, has eliminated ATM, a method of moving data and voice on fiber that enhances quality of service (QoS) but reduces IP transmission capacity. Instead, the company’s converged IP network runs directly on SONET, the framing scheme that overlays fiber, facilitated by intelligent multi-protocol label switching. The payoff: significantly greater speeds up to 10 Gbps on the backbone. “By placing QoS-enabled IP onto SONET we’re providing more bandwidth to our customers within the given fiber network that we have,” says Jim Krutchen, Onvoy’s manager of engineering.
CTC’s PowerPath Network currently incorporates ATM but will switch to IP on SONET by year’s end.
The fiber stops here
In most cases telecoms making the switch to IP-over-fiber depend on older, slower technologies to bridge the last-mile gap between DWDM backbones and their business customers. Generally, only companies housed in tall, shiny office towers in the downtowns of major-league cities enjoy direct-fiber hookups, usually from a Regional Bell operating company (RBOC) such as Verizon or Qwest Communications. Just 5 percent of U.S. commercial buildings tap into fiber, largely because of the high cost of running fiber under streets and parking lots: between $40 and $120 a linear foot, depending on the city. In outlying metro areas and smaller cities and towns, businesses plug into fiber arteries via coaxial cable or, more often, a copper T1 or SDSL line provisioned from an RBOC or other incumbent telephone company.
The carriers in the best position to roll out last-mile fiber on a broad front are the Bells (with the exception of financially ailing Qwest), but with many of their local-exchange competitors vanquished, they see little sense in junking copper infrastructure worth more than an estimated $1 trillion. “The problem is that [replacing copper] would be so enormously expensive that nobody’s going to do it for the foreseeable future, at least five years out,” says Veeral Shah, a telecom analyst with Battery Ventures, a bi-coastal venture capital firm.
Gigabit Ethernet (Gig-E) services offer an elegant workaround for telco data jam. Ethernet, the link-layer protocol that runs 95 percent of the world’s LANs, is tailor-made for racing data packets across long-haul or intra-city fiber networks. The catch is that Gig-E purveyors depend on fiber to deliver those packets to the customer’s door. That has restricted IP over Ethernet to urban cores, and even there, a number of metro Ethernet companies have failed to earn a profit. Yipes, a metro Gig-E provider based in San Francisco, made its name by selling Internet access and LAN-to-LAN connections as fast as 1Gbps to large enterprises. But in March the company filed for Chapter 11 bankruptcy protection, unable to generate enough revenue to pay for laying fiber to large office buildings in 20 cities. Another metro Ethernet company, XO Communications, went Chapter 11 in June.
Other IP-over-Ethernet providers such as Cogent Communications , Telseon, and IntelliSpace soldier on, convinced that there’s a business case for raw speed in Internet access and data transfer. David Schaeffer, CEO and founder of Washington, D.C.-based Cogent, doesn’t believe in telecom convergence: “We made a conscious decision to focus only on data services … in the future there will be data networks and voice networks, and they’re inherently very different.” Backed by a $410 million loan from Cisco, Cogent has built a 17,400-mile national and metro fiber network that puts packet-laden Gig-E on a stripped-down version of SONET. The system is fast and cheap; customers such as C-SPAN, the NFL, and inSORS, a Chicago-based video collaboration service, pay a flat rate of $1,000 per month for every 100Mbps of Internet access, up to 1Gbps.
But businesses in the suburbs or second-tier cities hankering for inexpensive, ultra high-speed Internet connectivity won’t get it from Cogent. The company, which has yet to post a profit, focuses on serving large multi-tenant buildings in the fiber-rich downtowns of big cities such as Los Angeles, Atlanta, Miami, Seattle, and Toronto.
Likewise, free space optics, a technology that gets around the high cost of last-mile fiber by doing away with fiber, is likely to remain a niche solution for companies in tightly packed downtowns. Laser systems from companies such as Terabeam, LightPointe , and AirFiber that relay data at up to 2.5Gbps between buildings require a clear line of sight, and can be disrupted by fog, rain, and high winds.
The telecom industry has a lot of work to do before IP on fiber displaces traditional circuit-switched and cable networks. In order for Ethernet data and converged voice and video to ride the light across the country and around the world, linkages must be forged among carriers that haven’t yet learned to speak pure IP. Making an end-to-end VoIP call from Chicago to New York or Madrid isn’t possible today, for example. To complete their journey, IP packets must enter PSTN and negotiate a series of circuit-switched handoffs between carriers deploying a hodgepodge of technologies. Not surprisingly, voice quality can suffer.
“The easy part is getting voice-onto-IP packets and moving them a long distance,” says Fazio of HeatSeeker. “The hard part is getting everybody to interconnect and find operational protocols and processes so that I can make a long-distance phone call that may traverse three or four different phone companies and just have it work automatically.”
Progress must continue in refining QoS standards, developing back-office systems for bandwidth self-provisioning, and dovetailing Ethernet with different flavors of SONET. Cross-network QoS should be easier to implement in IPv6, the emerging protocol expected to replace today’s standard for IP transmission. The recently introduced Generic Framing Procedure tackles the last issue, making it possible to transport any IP-friendly protocol, including Gig-E and Fibre Channel, over SONET.
Most important for small-to-medium-sized businesses (and eventually, consumers) eager to taste the fruits of the optical Net, RBOCs have to swallow hard and start converting the last mile to fiber, not just downtown but in suburbs and smaller cities. No current fixed- wireless or satellite technology can match fiber’s ability to deliver mega-bandwidth at low cost. Bells such as Verizon and BellSouth argue that passage of the Tauzin/Dingell bill, at press time awaiting action in the U.S. Senate, would spur last-mile investment by letting them charge local-exchange competitors market rates for local access. Opponents such as Cogent’s Schaeffer disagree, arguing that unleashing the Bells doesn’t guarantee that they’ll spend their newfound profits on metro fiber or other network upgrades.
Amid all this technological flux and market uncertainty, the vision of ultra high-speed Internet access on glass endures. Research is ongoing in next-generation technologies that promise to transport packetized data with greater ease than today’s cobbled-together optical networks, delivering boundless bandwidth for pennies a gigabyte.
OMNInet, a research effort at Northwestern University’s International Center for Advanced Internet Research (iCAIR), is putting leading-edge 10GB Ethernet technology to the test in applications such as deep data mining, high-resolution streamed video for telemedicine, 3D visualization for industrial design, and computational science. Essentially a mini-metro fiber network, the OMNInet testbed in the Chicago area uses advanced optical switching to move IP packets directly on DWDM. “There’s less equipment, and fewer layers and conversions, so it’s much more efficient,” says iCAIR Director Joe Mambretti. The experimental switches incorporate micro electro-mechanical systems, microscopic mirrors that route traffic on light waves.
Another iCAIR project, Starlight, seeks to create a worldwide all-optical network. And scientists at Bell Labs, the R&D arm of Lucent Technologies, have managed to push long-distance transmission speeds on DWDM to a mind-boggling 2.6 terabits per second.
The ultimate optical Net would be one in which there’s no electronic conversion at all; IP packets would piggyback on photons streaking along frictionless glass at the speed of light. You don’t have to be Einstein to grasp what such infinite bandwidth would mean for future iterations of the Internet. “That kind of network would be the fastest possible network,” Mambretti says. “It’s very much a research area, though; it’s on the far distant horizon.”
IPspeak: how to speak advanced internetworking
If Isaac Newton had studied fiber optics, that apple falling on his head would have just raised a painful bruise, not ushered in greater understanding. A dizzying array of framing mechanisms, transport protocols, routing devices, and network interfaces makes up the landscape of the optical Net, a landscape that constantly quakes and realigns itself as carriers develop their own models of next-generation fiber-optic networks. For those of us who associate the term ATM with fast cash, here’s a guide to IP-over-fiber terminology:
asynchronous transfer mode (ATM)
A data-link layer protocol that is being phased out by many carriers. ATM delivers excellent QoS for data, voice and video, but it doesn’t scale well above 600Mbps, and because of its small cell size, ATM imposes a “cell tax” of about 20 percent on Internet traffic.
dense wavelength-division multiplexing (DWDM)
A method of sending many different colors (wavelengths) of laser light down the same optical fiber at the same time, vastly increasing transmission capacity. Speed and flexibility-DWDM can simultaneously carry different types of signals, such as SONET and ATM-have made DWDM the de facto standard for fiber-optic networks.
free space optics (FSO)
Gigabit-speed optical connectivity sans fiber. Making a comeback after failing to live up to performance claims in the 1980s, FSO relies on lasers to beam packetized data in the terahertz spectrum to and from central nodes in urban areas. Because signal quality drops significantly below a half mile, the technology so far has proven most useful in tying nearby buildings into LANs.
gigabit ethernet (Gig-E)
A protocol derived from the standard underlying most of the world’s LANs that supports data transmission rates over 1Gbps on long-haul and metro networks. Because of its speed and relatively low implementation costs, many IT experts consider Gig-E superior to full-fledged SONET as a transport medium for data and Internet content.
integrated access device (IAD)
A router installed at a business that converts signals from standard phones into IP packets, which can then be combined with LAN, VPN, and Internet traffic on a converged IP network.
multiprotocol label switching (MPLS)
In an IP/MPLS network, routers forward IP packets based solely on the contents of an assigned label, not the destination IP address. This simplifies network management, allowing carriers to control QoS for a particular class of traffic, bypass network congestion, and create “IP tunnels” for encryption-free security on VPNs.
synchronous optical network (SONET)
A framing mechanism for fiber, originally developed for voice, that can also handle video, data, and Internet traffic. Almost all telecom companies deploy SONET in their networks; some have installed bare-bones versions (SONET-lite) built for speed in converged IP and Gig-E applications.
time-division multiplexing (TDM)
A form of multiplexing used to carry voice in traditional PSTN. Converged IP networks eliminate TDM-at least on the backbone-by transmitting phone conversations as well as Internet traffic as a stream of data packets.