Cable and other competitors to the local exchange carriers are at a crossroads when it comes to choosing how they will interface their networks with the networking world at large. In a nutshell, the question comes down to figuring what role, if any, ATM (asynchronous transfer mode) should play in managing local traffic. At a moment when the answers should be getting easier, they appear to be getting harder.
"There has been a religious-war-like quality to the discussion about ATM," said Mario Vecchi, senior vice president of network technology for Time Warner's Excalibur group. While the war is far from over, he added, the cable industry must approach its data network buildouts one stage at a time, taking care to ensure that, as "the bubbles continue to grow," the local data backbone linking hubs to headend servers and routers can handle the volume requirements.
This could well mean selecting a new technology like IP (Internet Protocol) switching or tag switching to support efficient interfacing of routers with ATM networks in the long-haul carrier domain, Vecchi said. But those technologies are still in flux, and, in many cases, operators need to make decisions now, especially when it comes to linking hubs and headends.
"There are decisions directly affecting scaling up the networks that are hard to make without knowing whether we're ultimately going to ATM," said another engineer at Time Warner, asking not to be named. "You can err on the side of conservatism and stay with FDDI (fiber data distribution interface) while the protocol stack is sorting itself out, but you risk running into Quality of Service problems if you get good market penetration of the service."
Many MSOs are avoiding making the choice by limiting their rollouts to single hub service areas. But it's likely they'll have to commit to interhub data trafficking techniques long before the ATM debate is resolved.
On one side of the war, telcos, which have driven ATM as their preferred design solution to marrying public switched and data networks since the late '80s, now believe ATM is in shooting range of becoming cost effective end-to-end. As a result, they are deepening their commitments to the cell-relay switching technique as they prepare to use DSL (digital subscriber line) technology to tap the mass market potential for data services.
But, at the same time, ATM is becoming a focal point of disappointment and contention within the data community, despite early successes in "edge" applications that distribute data traffic across multiple networks within a single corporation. "The ATM market is up in the air today," said Jennifer Pigg, vice president for data communications at The Yankee Group, one of many speakers noting this fact at the recent ComNet conference.
The issue of what role ATM is going to play in data communications has been center stage at major shows devoted to enterprise networking, where publicity surrounding new options like gigabit, IP and tag switching signal that ATM has yet to gain ascendancy in the datacom market. "This is the first year we've heard so much gloom and doom about ATM, which is especially noteworthy after so many years of upbeat predictions," said Frank Dzubeck, president of Communications Network Architects Inc., a consulting firm.
Pigg cited a failure of multimedia to take off in the corporate LAN (local area network) domain as one reason for the uncertainty. Moreover, the failure of carriers to push ATM as a voice as well as data switch, as opposed to being a mere aggregator of combined traffic or a router of data traffic only, has hampered its usefulness as an integrator of circuit voice and data traffic at the LAN level of business operations.
"Voice is not a good idea over ATM at this point," Pigg said, adding that this will change once next-generation ATM protocols are resolved, and carriers implement the advances in new switches. "Carriers are rethinking their approaches to solutions for data traffic at the edge, in the core and on the global level," she said.
Many people charged with building national data networks believe there are means and the facilities in place to exploit them that can meet their trafficking requirements cost-effectively without moving data through switches. As a result, they contend the business end-user market will not be drawn to ATM equipment, choosing instead to build on the embedded base of Ethernet and other widely-used data network protocols. This, in turn, will make ATM unnecessary for the consumer market as well, they say.
"I predict we never will see ATM as an end-to-end solution," said Milo Medin, senior vice president of technology for @Home Networks, speaking at a recent conference on bandwidth solutions in San Francisco. "Fast Ethernet is eating ATM's lunch in the enterprise market, and gigabit Ethernet is going to have it for dessert."
Coming from a man who, two years ago, spoke exuberantly of building a national ATM network to bypass Internet bottlenecks, Medin's comments suggest the new infrastructure design ideas sweeping through the data networking community may be cause for serious rethinking of migration strategies in cable. But drawing too firm a line right now based on current falterings of ATM risks ignoring several significant trends, including the growing support for ATM as an end-to-end solution in the telephone industry, as well as technical solutions that promise to make ATM a better choice over the next year or two.
Clearly, ATM, as a vehicle through which the telephone companies draw the data transport business of the enterprise sector, is already a pretty good working engine offering cost-competitive alternatives to private networks in the wide area backbone, which is why analysts have given it strong support. But that doesn't mean it's a viable solution for most companies, said Christine Heckers, vice president for broadband services at TeleChoice Inc., a consulting and research firm.
"The ATM services that are out there work very well and are affordably priced in terms of carrier rates and costs of equipment," Heckers said in a ComNet presentation to business managers. "But unless you are prepared to make a very strong investment in human resources and time on your staff to train people to adapt to ATM, you should stay away from ATM, this year and probably next."
That's a long way from saying "never" to ATM to the desktop. Or to the home, especially if, as Nynex, Bell Atlantic, SBC Communications, Ameritech and other telcos believe, the cost curve for ATM to the home is pointing toward deployment within the next two years, possibly as early as a year from now.
Those who claim ATM end-to-end will not be practical anytime soon "are operating on outdated assumptions," said Tom Eames, co-president of General Instrument Corp.'s NextLevel Communications, which is soon to be the name of the reconstituted and spunoff GI Communications unit. NextLevel recently was named supplier for Nynex's planned deployment of switched fiber networks, which envisions use of ATM all the way to residential premises.
"When you get down to 0.5 and 0.35 micron silicon technology, the cost penalty for deploying ATM all the way to the customer premises goes away," Eames said. Within a year, he added, the circuitry in the ATM chips will be reduced to 0.25 micron dimensions, which will cut chip costs by a factor of four.
"What was once a $40 problem becomes a $2 or $3 problem, and, in two years or so, becomes a 50-cent problem, which is no problem at all," Eames said.
Like Bell Atlantic, which recently announced plans to build switched fiber networks starting in Philadelphia this year, Nynex plans to position itself for delivering broadband services by installing fiber loop systems that derive immediate cost benefits on the telephony side, starting in the Boston area by the end of the year. Assuming the technology performs as expected, Nynex will be in a position to easily add video and high-speed data services by inserting ATM circuit cards at the broadband digital terminal in the central offices and at the ONUs (optical network units) in the field, officials said.
But Nynex has not issued an RFP for ATM switches, which would be required to support the move to full service over the fiber networks, said Walter Silvia, vice president of broadband at Nynex. He noted the company would also have to select a video software system for supply of that type of service over its fiber networks.
"We don't have a (broadband services) market trial planned at this point, but we'd expect that, over time, we would conduct such a trial before moving forward," he added.
The facilities topology used in the NLC system entails extension of fiber to ONUs serving eight or 16 subscribers, with a hybrid drop consisting of a coaxial cable and two twisted-pair copper lines extending to each household or business from the ONU. In the broadband mode, signals are delivered to the ONU in a 155 megabit-per-second (OC-3) Sonet stream from the broadband digital terminal (or host digital terminal, as it is more commonly referred to), which can be located at the central office or in a remote terminal or hub.
For broadband applications, signals are all mapped to the 53-byte cell of ATM and converted at the set-top, telephone or PC interface to the native device protocol, Eames said. "We load the TDM (time division multiplexed) traffic as well as data and digital video into the ATM cells and broadcast them to all devices in the home," he explained. The ATM throughput is based on the 51.84 Mbps data rate endorsed by DAVIC (Digital Audio Video Council), he added.
NLC plans to make available an ATM-based network interface card for PCs to avoid the process of converting the ATM signal to Ethernet protocol, noted Peter Keeler, NLC's other co-president. "We dedicate up to 19 Mbps per customer in the upstream," he added.
Given the telephone industry's recent swings from one network topology to another, there's no certainty that, this time, the plans will be executed as described, although Bell Atlantic officials reported construction on that carrier's FTTC installation in Philadelphia has begun. But the consistent theme, no matter which topology is the flavor of the day, remains ATM.
For example, while FTTC deployment is barely getting started, DSL (digital subscriber line) technology, delivering high-speed data at various rates over twisted-pair copper lines, has a shorter time-line and a bigger head of steam at this point, with most carriers saying they'll begin offering commercial DSL services this year. And, in many cases, the transport format of choice is ATM.
The universality of the ATM factor is much more important to the pace of telco broadband rollout than the differences among topologies, noted a consultant working with americast, the video venture involving Ameritech, BellSouth, SBC Communications, GTE Corp. and SNET Corp. "If you look at what the partners are doing (from a facilities standpoint), you can say they're at odds with themselves with a lot of different strategies," the consultant said, asking not to be named. "But ATM is the way they get to cross-platform scales of efficiency in terms of overall integration and service profiles, and we feel that's moving ahead very well."
Perhaps the biggest threat to ATM is the emergence of IP switching schemes that would allow data networks to interface efficiently with ATM-switched networks without forcing users to abandon the cost efficiencies of using IP to build functionality into their Intranets. IP and tag switching, developed, respectively, by Ipsilon Networks Inc. and Cisco Systems Inc., use simple protocols to identify and dynamically find long flows and put them in virtual circuits in front of an ATM switch, thereby exploiting ATM backbone efficiencies while preventing penetration of ATM beyond the private network edge.
"The hardware of an ATM switch is about 10 times faster in terms of just raw bits through the box as it is for a router, typically at perhaps half, down to a fifth of the price, depending on the features you want," said Larry Lang, vice president for product management at Ipsilon Networks Inc., the first to introduce IP switching. "Unfortunately, that ATM hardware is buried under a lot of ATM software with a lot of jargon attached."
Cisco Systems' tag switching variation on the IP switching theme is especially appealing to many cable operators, given the substantial penetration of Cisco 7500 routers in the cable data networking infrastructure. The system uses message tags rather than direct insertions into the IP header to identify flows and put them on virtual circuits at the router. As with IP switching, this avoids the lookup process at other routing points while allowing ATM switches to participate at the layer 3 routing level.
"Tag switching will allow us to do edge-based source routing, which allows us to use IP and ATM in a single network much more intelligently," said Paul Bosco, director of broadband infrastructure and service development at Continental Cablevision. "By some time (this year), we'll have to get beyond the DS-3 limitation at the edge of our network, which means moving to the next generation router with or without ATM."
But tag switching, like the other proposed solutions, is in flux, with vendors at loggerheads over achieving the standardization that is essential to widescale market acceptance. In February, IBM, 3Com Corp. and Cascade Communications lined up behind Ipsilon's IP switching in an effort to thwart Cisco, which could lose its dominance of the router business if IP is chosen over tag switching.
And even greater uncertainty surrounds gigabit Ethernet switching, which, unlike options like IP and tag switching, would supplant ATM altogether with devices that match ATM's raw throughput speed while retaining the native Ethernet format in the backbone. Gigabit switches won't be available until late this year at the earliest, and another great standards debate looms before the market will see an interoperable multivendor solution.
It's true that "ATM standards are moving slower than a senatorial debate on C-Span," acknowledged Roger Kosak, industry solutions project manager for IBM and chairman of the ATM Forum's network management working group. But, he added, that's because getting software to do all the things everybody is looking for in ways that meet everyone's specific requirements is tedious.
"There are still a bunch of competing proprietary solutions out there," Kosak said, noting that IBM's backing for IP switching in no way represents a loss of support for all-ATM solutions. "I'd submit to you that for (IP switching) to be a viable technology, it's going to take a little time to figure out how to really do Quality of Service and make it work right for all the people that want to implement the technology."
Telcos are gambling the headstart for ATM and the mass market base they can build to drive chip costs ever lower will translate into an end-to-end broadband solution that is ideal for their networking environment. Betting against them and the use of ATM in the cable modem environment translates into a bet on the ability of the datacom world to standardize on alternate solutions.
It may be a good bet, but it's not as sure as some data pros might make it out to be.