Transport Truce

In the battle for converged-network dominance, old rivals ATM and IP will coexist—for now.

By Margot Suydam, Technology Editor

Internet protocol (IP) has become the pie in the sky for the converged voice/video/data network, promising ubiquitous end-to-end connectivity for the transport of communication, entertainment, and information. As with every pipe dream, however, the reality often hits early and hard. There are hurdles to overcome and market and technical issues to address before the dream becomes a reality.

And so, the journey toward nirvana may begin with a big leap forward, but is often followed by a fall backwards. And ultimately the process ends up being nothing but a series of baby steps to actualization.

Such is the row that IP has sowed on its way from technology of choice for the Internet to the promised land—providing a converged network offering. No doubt, the buzz around Voice over IP (VoIP) has reached full tilt as a replacement for the circuit-based public switched telephone network (PSTN). Meanwhile, CLECs (competitive local exchange carriers) are looking to enter the local telco market with voice-over-DSL services, while ILECs (incumbent local exchange carriers) and CLECs alike aim to bundle video, voice, and data to compete with the cable TV industry. But the reality is that while IP may be the data format of choice, the transport protocol for most of these new broadband networks continues to be ATM (asynchronous transfer mode).

Long-haul history

ATM, a technology based on transferring data in cells or packets of a fixed size, is known for ensuring quality of service (QoS). ATM's small, constant cell size allows the transmission of video, audio, and data over the same network, while assuring that no single type of data hogs the line. As a result, service providers can guarantee to their customers that end-to-end latency will not exceed a specified level. Such a concept contrasts with the best-effort service model that is pervasive on the Internet. While QoS enhancements for IP are in the works, the consensus of many is that IP is a long way from putting ATM down for the count.

"ATM has been reinventing itself for the last 12 years," says Steve Perna, vice president and general manager of the Optical Networking Division of semiconductor provider PMC-Sierra. "While everybody has taken pot shots at it, ATM has evolved, primarily because the things it does well, it does better than any other protocol. People have woken up to the fact that ATM, originally perceived as the technology that would make broadband ISDN happen, in reality was designed to add determinism to the network and to bring QoS to what was largely a constant-bit-rate network. When you constrain ATM to those requirements, it does a good job."

Originally intended as the protocol to haul premium services, ATM has had a specialty role providing transport of voice and video. "ATM, SONET/SDH, packet over SONET, Gigabit Ethernet physical layers, and frame relay—all of these are services or protocols that equate back to what users are interested in, such as sending a long file or getting full-motion video on an Internet feed," Perna says. "In the early '90s, people were talking about putting ATM adapters to replace Ethernet on the desktop, ultimately putting it on the motherboard. It never happened, because ATM was a premium service for a premium application. People found that they didn't need all of that on the back of their PCs."

Over the years, ATM got back to its roots as a WAN technology, and found wide use in the core of the network, the edge or metro-area network, and the access network. More recently, however, the network core has increased in speed and bandwidth, enabled by an emerging plethora of optical-based technologies. "ATM is not in the core of the network anymore, but is much closer to the edge or metro," Perna says. "The metro network is now where there are a lot of the servers, and so a lot of the intelligent decisions are made there. The core, meanwhile, has evolved to be unsophisticated and unintelligent, but high performance in point-to-point optical transport applications. ATM is now seen largely as a service in the metro and as a transport technology in the access region."

Access attributes

In the access network, most DSLAMs (DSL access multiplexers) use ATM to carry voice and data over the last mile, because ATM solves the local-loop QoS issues much more easily than IP. In fact, the ADSL Forum has specified ATM for transport of data to the customer premise equipment (CPE) over ADSL. "Carriers don't want to obsolete themselves," Perna says. "They want to start layering services onto a voice-based business, because that is what makes money. Voice is still the large revenue driver in networks, while data or IP networks [represent] by far the largest incremental increase in traffic."

CLECs have tried to crack the local telco market with IP-based services, but facing a monopolistic voice business dominated by the incumbents, many of them failed. "The only way CLECs could differentiate themselves was to come in with an IP-based network that had as its business model voice over packets—with essential voice for free and charge for packets," Perna offers. Free phone calls over the Internet have not really materialized yet because IP networks inherently support large, variable, and "bursty" file transfers—not conditions that are amenable to voice traffic. "The other major mandate is to provide for very robust and reliable connections," Perna says. "Anything can happen and you still get a dial tone."

While companies like Nortel have tried to drive the concept of "Web tone" on IP-based networks to approach the reliability of circuit-switched networks, according to Perna, ATM QoS was what the carriers wanted to extend their very reliable and lucrative voice business to a multiservice Internet business.

"ATM QoS is the guts behind service-level agreements for the carriers as they provide current and future services," he says. Those with a need for high-priority traffic can pay a premium to get it. Those with less demanding needs can save money. "IP is now a service offered along with ATM as part of a whole multiservice package," Perna says. "What's prevented IP from being ubiquitous in these networks is that IP is still not adequate as a transport mechanism. IP is geared for high performance, speed, and the variable bit rate of email packets."

Louise Gaulin, voice-over-ATM product-line manager at Mitel Semiconductor, concurs that due to QoS issues, ATM makes more sense than IP when delivering services, whether broadband or wireless voice. "Today, in applications like voice over DSL and 3G wireless, vendors are pushing back on the idea of an end-to-end IP network." They're finding that the standards are not there yet and that IP QoS remains unproven. "They know that ATM has good QoS and they want to minimize the risk," she continues. "The 3G infrastructure, switching architecture, and traffic management is ATM, but at the access network they convert to go out as IP. In voice over DSL, meanwhile, there are companies offering ATM and IP IADs [integrated access devices]."

Mitel offers the MT90502, an AAL 2 (ATM adaptation layer 2) device that enables media gateways or multiservice switches to encapsulate voice at a variable bit rate for transmission over ATM. "In AAL 2, you have smaller packets that can vary in size in a cell which is a fixed length, making it suitable for a converged voice and data network," Gaulin says.

Bandwidth is the primary justification for the QoS mandate in the access network, asserts Chung Ho, assistant general manager for the NEC Electronics Communications Division. "On the CPE side, with data and voice integration, you're living with a variety of limited-bandwidth technologies, at least for now." The different flavors of broadband—DSL, cable, and satellite—all have limited bandwidth, which consumers and businesses will eventually consume. "Once that is filled, you have to worry about QoS," Ho says. "You want to give preferences to different types of traffic."

NEC Electronics provides a family of network-controller chip sets for the IAD and CPE markets. "ATM still has a lot of life, being able to provide differentiation," Ho continues. "If you want to provide telephony, you don't want voice calls to be dropped. Our network controllers support ATM, so we can do IP. ATM has more demand on the hardware side, while IP is inherently a higher-layer function."

While Ben Vos, director of network planning and design with Sprint, acknowledges that the word "debate" is attached to a discussion of IP and ATM, there is no such debate within his company. "There are real values and benefits for each protocol at the moment, and going into at least the next two years," he says. "Certainly, we see things changing as IP and things like MLPS are developed that start to provide some of the features and capabilities of ATM."

MPLS (multiprotocol label switching) is an IETF (Internet Engineering Task Force) initiative that enables QoS by integrating Layer 2 information about network links (bandwidth, latency, utilization) into Layer 3 (IP) within a particular autonomous system—or ISP. (For more on what MPLS brings to the party, see the sidebar, "MPLS emerging.")

ATM technologies form the basis for the access portion of the Sprint network and its ION offering. "When there is a need for multiple service types and service characteristics maintained though the access piece of the network, you want to be able to dynamically assign bandwidth to different applications, yet guarantee a certain level of performance or QoS for different connections," Vos says. "ATM is very well suited for a multiservice access environment, especially with DSL technology. And that is where the market is going."

"As you get to the point where you provide the first level of switching or routing for those services and then go into the backbone of the network, today we're supporting multiple types of connectivity," Vos adds. "If it is strictly IP best-effort service, we absolutely want to hand it off as an IP-over-SONET type of connection. We will give up, at least at the moment, that QoS capability across that IP backbone. We also maintain the ability to route over the ATM infrastructure for those services that do require specific QoS." Sprint supports both regimes today, and Vos says that will continue for the next couple of years.

Standards are essential because they allow Sprint (or any provider) to choose equipment from multiple vendors, picking the best-in-class product for each different network element. "Standards give us the assurance we are not painting ourselves into a corner in terms of proprietary technology," Vos says.

Peaceful coexistence

Others agree that ATM and IP are complementary network technologies. "I don't see it as an ATM-vs-IP scenario," says Bruce A Kulp, senior product marketing manager of data services at Global Crossing, a provider of services over a global IP-based fiber-optic network. "We layer ATM as well as Frame Relay on our IP network."

Global Crossing recently lit up its voice-over-IP backbone, and has core VoIP gateways in 22 cities worldwide. Also, its customers can leverage its high-speed ATM network to support multiple applications—data, Internet, voice, and video—over a single platform. Global Crossing's ATM network carries directly over DWDM (dense wavelength division multiplexing) and supports multiple classes of service.

"ATM technology allows users to prioritize their traffic and allocate capacity to support multiple services," Kulp says. "By consolidating various services over a single delivery system, ATM provides better economies of scale by reducing operational costs, improving network efficiencies, and simplifying management needs."

Moreover, Kulp contends that ATM has migrated to the edge of the network because the cost of the technology has gone down. "The intelligence of the network is moving away from the core to the access network," he says. "And so we've seen the emergence of a market for ATM-based DSLAMs and IADs. That's happening because ATM, which was once limited to priority service, has become less expensive to deploy."

According to Tim Thompson, marketing director of Mitel Semiconductor's Packet Switching business unit, ATM and IP will coexist for some time to come in related but different markets. "For the next five to 10 years, the market for each technology will be unique from the other," Thompson says. "Because service providers do not want to do a forklift and strip everything out, it will be a slow and transitional process before IP is inserted into the network with the ability to interoperate and communicate either through native ATM or using MPLS."

Mitel offers the MVTX2600 family of VoIP switching ICs, which consists of three single-chip VoIP switches that support the bandwidth and QoS requirements of voice, video, and data traffic in Ethernet LANs. Mitel is targeting vendors that provide equipment for the enterprise as well as multitenant/multidwelling units (MDUs and MTUs).

"The issue is, how far out can packet-switching technology go and in what period of time," Thompson says. "There are companies doing IP-based DSLAMs, but they are unique. The MTU/MDU market is a perfect example of an area where you can get away with that type of application. There are boxes on either side, so it doesn't matter what goes between them. You can have islands of pure IP, and the QoS we are developing is comparable to ATM but is not mainstream nor standardized.

IP proponents

Despite the momentum behind ATM in the access network, a number of companies remain firmly committed to IP as a viable alternative. One such company is Net to Net Technologies, which provides simple, low-cost, packet-based xDSL products optimized for IP transport.

"Broadband access was not enabling broadband access to a lot of people," says Ken Latimer, the company's president. "Everyone is doing ATM DSLAMs. I've been involved with ATM back to when it was going to replace Ethernet to the desktop. So I thought about what had happened to the LAN world and tried to relate it to this one. Ethernet is just so straightforward. It's easy to configure and create new versions, and it's competitive, so people can create new products. ATM is very configuration intensive; it has to be right or things don't work."

AccessLan Communications, a supplier of DSLAM equipment for Voice-over-DSL, also comes down in favor of IP-based access. "The move to IP has already yielded substantial benefits at the core of the network, fostering revolutionary innovations in optical switching and transmission," says Byron Young, AccessLan's vice president of product marketing. "The optical core and the enterprise LAN are already IP-based, but access service providers have been hindered by a lack of access-ready convergence solutions that give them the same cost and service advantages as core service providers."

AccessLan's i-SLAM architecture provides a robust access infrastructure for IP-based services, according to Young. "Features like IP Services Aggregation (IPSA) eliminate the need to provision virtual circuits all the way from customers to switching centers, allowing the network to handle the scalability issues associated with the tens of thousands of users anticipated with mass deployment," Young says. "Also, QoS, like MPLS and Label Edge Routing, at the point of access can ensure the delivery of reliable IP-telephony services."

Another IP cheerleader, Integral Access, provides a multiservice packet-based broadband access system with the aim of enabling competitive carriers to offer differentiated services and combine voice, video, and data on new generation IP-based networks as well as traditional circuit-switched infrastructures. Using MPLS and traffic management, PurePacket delivers QoS and enables service providers to deliver, manage, and bill for IP services and applications.

"Whether DSL or T1, and fiber or wireless in the future, IP is the play everywhere, and MPLS is the key technology to achieve end-to-end QoS," says Guy Chenard, Integral Access's vice president of marketing. He asserts that ATM's days are numbered. "ATM was believed to go right down to the desktop and it didn't," he says. "Designed before the Internet revolution, ATM will not scale long term for the type of traffic and the way we'll deliver traffic. The core will be composed of essentially fiber-based DWDM of which services will ride on an IP MPLS router base. That is the most cost-effective way to transport IP traffic through carrier-class systems."

Integral Access uses MPLS to manage the limited bandwidth between the network and the CPE. "If you are using DSL or a T1 line, it's very easy to oversubscribe, so we make sure that mission-critical services are dealt with," Chenard adds. "This allows service providers to offer differentiated services. They can set a threshold and bill accordingly."

When asked if carriers are willing to trust IP-based services, Chenard comments, "Most carriers will say 'prove it to me that IP is the way to go.' We say most of the technology is ready to build a voice-over-IP network. What's missing is the full set of SoftSwitch features. That will take six months to a year before it is all there, and after that pretty much everything is there. Also MPLS, from the access to the core, needs to be demonstrated. What we see is that carriers want to deploy IP into a new environment or market. They get their arms around the technology and then they move from market to market.

MPLS emerging

With the arrival of MPLS-enabled equipment from companies like Cisco Systems, emerging network service providers can reap the benefits of end-to-end optically based IP networks. IP enables value-added services, while optical adds bandwidth and lowers costs.

One company grabbing onto these advantages is Cambrian Communications. Cambrian is constructing an all-optical, fiber-rich metro and long-haul IP network using Cisco's IP+Optical offering. The company aims to build an advanced, next-generation Internet infrastructure capable of delivering cost-effective transport and multiservice offerings in metropolitan area networks (MANs) along its backbone connecting New York, Philadelphia, Baltimore, and Washington, DC.

Cambrian plans to meet the wholesale communications needs of emerging service providers, carriers, and major private network owners. Through the combination of IP and MPLS and its high-speed optical backbone, Cambrian can deliver a full spectrum of differentiated services.

"We want to be the transport network of choice," says Joe Cecin, Cambrian's COO. "It doesn't matter what our customers are putting on to the network. They're probably doing a combination of voice and data aggregation. For example, there are DSL companies co-located in different central offices. They do traffic aggregation with unicopper pairs, but then typically they have no network infrastructure to back traffic out with, so they're dependent on the incumbents to provision services out of those markets. It's expensive and has long provisioning times. They're looking to get more control."

Cambrian chose Cisco's IP+Optical equipment because it supports MPLS and allows the company to ensure QoS. "Historically, networks were designed to be a voice transport, with data as a sideline," Cecin says. "The world has switched on us. There is more data being transported. And so ATM is less efficient. We address the IP QoS issue with MPLS—what a lot of people refer to as ATM on steroids. It allows you to send large packets when latency is not an issue, but at the same time define latency-sensitive traffic in a way that you can get it through your network in a timely manner."