It's IP vs. ATM in all-out-war
Practitioners of the anything-but-ATM data networking religion are finding ever more reasons to believe IP-based transport solutions represent a winning approach to taking on incumbent carriers, but they shouldn't get too carried away.
Even as the rapidly expanding power of IP (Internet protocol) Layer 3-based architectures opens unlimited service potential at unprecedented levels of cost efficiency for next-generation providers of broadband voice and data services, the incumbent local and interexchange carriers who comprise the core of the ATM (asynchronous transfer mode) camp are finding ways to have their cake and eat it, too — which is to say, to stay with the legacy-network integration power of ATM while opening an evolutionary path to an ever-more IP-dominated infrastructure.
For example, while Sprint, with its recently-announced Integrated On-Demand Network (ION) local access strategy, says it will install an ATM gateway box in the homes and offices of all its local service customers, the carrier will not be forced to supplant Layer 3 IP with ATM Layer 2-based functionality when IP-based services are streaming through the gateway, says Steve Hratko, business development manager for Cisco Systems, the key supplier in Sprint's ION strategy. "The Layer 2 device at the customer premises is transparent to the router upstream," he says.
In essence, the technical versatility of new IP-friendly ATM products entering the market from Cisco, as well as traditional telecom suppliers like Alcatel, Lucent Technologies, Nortel, Siemens and Ericsson, narrows the efficiency gap between established carriers and new ones that would use an all-IP infrastructure to support low-cost entry into the incumbents' markets. The gap is still there, but so are the advantages of incumbency, making the new entrants' efforts to challenge major carriers a riskier venture than it seemed just a few months ago.
"Maybe if these incumbent carriers were moving to broadband services two or three years from now, they'd take a more IP-centric approach," notes Kevin DeNuccio, vice president for service provider operations at Cisco. While, in the future, integration of legacy systems might be doable through IP rather than ATM gateways, "at this point in time, carriers have to ask themselves whether the QoS (quality of service) that ATM supplies is going to be there in IP gateways when they flip the switch," DeNuccio says.
While QoS over IP, supporting such ATM-touted capabilities as bandwidth on demand and guaranteed bitrates, is a fast-moving force that offers great promise for cable MSOs and other would-be entrants into IP-based services, it isn't there on a scale sufficient to permit translation of all of a major incumbent carrier's legacy traffic onto IP backbones, even in the eyes of someone as committed to such backbones as Steven Blumenthal, vice president and general manager for GTE Internetworking's global networking infrastructure. "As we're building our new (national) fiber backbone, initially we'll have separate circuits for voice, ATM and frame relay," Blumenthal says." But over the next two to three years, you'll see all that converging (to IP)."
GTE is opening a nationwide path to this convergence by exploiting the power of high-speed routers to deliver IP traffic directly into a Sonet multiplexer for transport across the backbone. "We're relying on the router's ability to deliver class of service and quality of service, to be able to mix real-time traffic like voice, video and multimedia and to give that an express lane through the network while mixing that traffic with lower-priority data, all at the IP layer," Blumenthal says.IP's attractiveness
IP, or, as it is officially known, packet-over-Sonet technology, has given network service providers of all stripes a powerful tool in the battle for maximum efficiency. By using next-generation routers from Cisco that currently feed directly into OC-3 (155 megabits per second) and OC-12 (620 Mbps) Sonet links and, by the end of the year, will be feeding OC-48s (2.4 gigabits per second), GTE is avoiding a costly "cell tax" that ATM switches imposes on IP traffic, Blumenthal notes.
This loss of efficiency results from the fact that fitting variable-length packets into 53-byte ATM cells often leaves ATM cells only partially filled with information. In addition, with five bytes devoted to the header in each ATM cell, there's significant redundancy of Layer 3 header information in an ATM cell stream.
The cell tax, based on in-house lab tests of ATM switches from a variety of vendors, turns out to be about 17 percent of capacity for GTE's mix of traffic, which consists of IP packets in the range of 64 bytes in length and big Ethernet packets of 1,500 bytes. "At the time we did the tests, we were leasing very expensive circuits across the country and overseas, and we couldn't afford to waste this bandwidth," Blumenthal says.
The implications of what GTE and other forward-thinking network service companies are doing with IP technology apply to the management and trafficking of signals at the local level, as well as across the backbone. For example, America Online is moving to use IP over Sonet in conjunction with advanced routing at Layer 3 in its regional data center networks after discovering the technology need not be limited to the backbone, says Victor Parente, chief network architect at the service provider.
"When I started looking at IP-over-Sonet technology, what we were looking for was a replacement for the current DS-3 (45 Mbps) technology and ATM that we were using to aggregate inter-site data center traffic," Parente notes." There are a lot of problems with (the ATM approach) in terms of the lack of maturity of the software, and also in the overlap or inconsistency between routed IP architectures and switch architectures."
The company decided to tear down the ATM complexes, choosing instead to operate "clear channel" connections between the centers using IP-over-Sonet. "All that complexity was replaced by a simple pair of routers running back-to-back," Parente says. "Over time, we also saw the technology as useful in our peering networks with our dominant Internet providers, and, finally, we've also gotten to the point where we're using IP-over-Sonet really as a data center technology, replacing LAN technology."
AOL is wrestling with issues that other providers of mass market services, such as high-speed cable and ADSL providers, must deal with as they ramp up to millions of customers. "How do I bring in many gigabits of Internet capacity into a data center?" Parente asks. "How do I design a new network that's going to allow AOL to peer with multiple ISPs?"
AOL must handle in excess of 10 gigabits per second per data center and sustain high port densities connecting up to 64 routers, Parente notes. "We needed redundancy and reliability, and what I call LAN/WAN transparency, which I think is one of the really critical values of IP-over-Sonet," he says. "This means that when I'm moving bits from one computer room or data center to another, I want to be able to take that data transfer on a continuous path across the road or across the country without changing data rates and protocols."
Circuit management is made easier as well, Parente says, noting that the ATM mesh array, to fit the size of AOL today, would require troubleshooting 1,200 circuits, as opposed to the eight circuits he has to watch using IP-over-Sonet. Parente says he also looked at gigabit Ethernet but decided that "doesn't scale very well." A gigabit may sound like a lot, he adds, "but my job is to plan further into the future, and IP-over-Sonet scales a little faster."
Indeed, suppliers are primed to push aggressively ahead with ever-higher capacity IP-over-Sonet solutions, says Graeme Fraser, vice president of marketing at Cisco. "By the second half of 1999, we'll see a transition to OC-192 (10 Gbps) based on packet-over-Sonet," he says. "Today, this is a point-to-point technology, but there will be an evolution to more complex architectures, starting with rings and then moving to mesh topologies that use Layer 3 routing (at all OC levels)."Road Runner's application
Actually, the evolution to ring topologies has already begun. For example, Time Warner's Road Runner Group is using IP-over-Sonet rings as its regional backbone in upstate New York, says Mario Vecchi, CTO for Road Runner. With Road Runner and US West Media Group's MediaOne Express data service about to merge, the new infrastructure going into place in New York is likely to be a template for the national infrastructure as a whole, Vecchi adds.
The new infrastructure model, setting a facilities-based line of demarcation between the network operating responsibilities of the local cable system and those of the national data provider, is being used to support launches this summer in Syracuse and Rochester. The regional system employs dual OC-48 Sonet rings supplied by Fujitsu Network Systems together with gigabit switch routers from Cisco to provide IP-over-Sonet connectivity between the two cities and, eventually, to Binghamton, where high-speed services are already in operation.
This regional architecture allows Time Warner to layer the network so that Road Runner can "assume responsibility for systems, applications, IP networking and a portion of the transport," Vecchi says. "This structure creates a hierarchy that separates functions and keeps protocols isolated as we build a total system."
Sprint, too, is at the cutting edge of deploying IP-over-Sonet rings, says Vab Goel, principle network design engineer at Sprint's Internet Engineering unit. "Packet over Sonet scales from 155 megabits (per second) to 2.4 gigabits," Goel says. "Two or three years ago, people thought that to move data that fast you needed ATM or gigabit Ethernet, but that's no longer true."
Sprint is taking advantage of existing four-ring Sonet networks already in place at the regional level to support a fault-protected international IP network that avoids the overhead penalties of ATM and Ethernet, Goel adds. "If there's a fiber cut, within less than 50 milliseconds the backbone will switch to a protection channel," he says. "No one will see any packets dropped."
On the surface, Sprint's IP-over-Sonet strategy seems to be in direct conflict with the heavy stress officials put on ATM in announcing their ION strategy in early June. But the conflict appears to be more a matter of confusion among executives describing the ION than it is a question of any internal split over ATM versus IP.
In describing how households would be able to use the carrier's ADSL (asymmetric digital subscriber line) links to generate services on demand, executives asserted ATM would be used from the premises to accommodate voice as well as data and video. But ADSL systems such as Cisco is supplying to Sprint are designed to carry circuit-switched POTS (plain old telephone service) over the same line with the ATM data flow, avoiding the costly process of converting telephony signals at the premises into and out of the ATM format.
What the executives appeared not to grasp was that their strategy, calling for bypass of circuit switches altogether, doesn't require a translation of analog circuit to ATM data at every home, which would be extremely expensive. Instead, says Cisco's DeNuccio, the switch bypass can be done by aggregating POTS circuits at a local IP telephony gateway server and then passing those signals directly into the backbone.
"My take at the high level is that this is a significant migration from the circuit switch to packet-based implementation of IP at the edge of the network,"DeNuccio says. Where circuit-voice/ATM integration will be cost-effective, he adds, is in the enterprise environment where multiple voice lines feeding a PBX (private branch exchange) are combined with data at the ATM gateway.
Whether signals are coming into the edge switch from end user premises or going out over the backbone as ATM or IP data makes no difference to the switch Cisco is supplying for Sprint's ION, DuNiccio says. In fact, he adds, the switch, employing Cisco-pioneered tag switching that is now part of the emerging MPLS (Multiprotocol Layer Switching) standard, supports a much fuller integration of IP and ATM than previous iterations of tag switching, which use a simple label attached to an IP packet to tell an ATM switch or IP router what priority to assign a packet without requiring the switch to search the IP header for class of service information.
"We use the IP intelligence layer to create a directory-enabled network that recognizes who the end users are and what services they are authorized to receive," DeNuccio says. "The terminology distinguishing ATM from IP is getting cloudy as we move to this new level of integration."MPLS: What is it?
While work remains to be done to finalize MPLS as a standard, Alcatel in early May became the first traditional telecom switch supplier to announce it was moving ahead with implementation of Cisco's version of the technology. "As carriers begin to take data onto their backbones, MPLS is an efficient, scalable way to support quality of service," says Edward Kennedy, vice president of marketing at Alcatel's Enterprise and Data Networks Division.
In effect, MPLS allows big carriers with core ATM switches to come as close as possible to the pure Layer 3 connectionless network model without reconstituting their entire architectures, Kennedy says. "Carriers want to retain ATM because they risk having to over-provision (bandwidth in) their network to accommodate a purely connectionless environment, plus there are issues of troubleshooting and billing control to be resolved in the IP domain," he adds.
Kenneth Frank, research director for BellSouth's science and technology group, agrees. "ATM has a lot of momentum to get at QoS, and trying to replicate what's been done in that domain with IP isn't an appealing prospect for us," he says.
But, as BellSouth rolls out ADSL service across its territories, starting with New Orleans next month (August), maintaining transparency for the Layer-3 QoS and other applications built into IP is an important part of the strategy, Frank notes. In fact, he adds, QoS won't even be implemented over ATM in the initial rollouts of ADSL, which gives the carrier flexibility to size up options later on, as QoS continues to mature in both the ATM and IP arenas.
Further adding to incumbent carriers' flexibility to exploit advantages in the IP domain is the imminent introduction of a new generation of carrier-class IP voice gateways, which are servers that interface IP voice with circuit-switched networks. "We're seeing implementations of gateway servers with capacity into the high 90s to 125 (DS-0) ports this year, and going well beyond that level in the future," says Michael Cassin, product manager for IP voice services at Concentric Network Corp., a Cupertino, Calif.-based ISP that is part of the global InterLine consortium of IP voice providers. "And we're seeing servers that are starting to be enabled with the backend system APIs (applications program interfaces) that are required for carriers to implement this stuff in networks."
For example, Lucent Technologies, already in the market with its own IP telephony gateway system, is moving on several other fronts to support the mounting carrier demand, even going so far as to bring out a 128 gigabit-per-second IP switch this quarter (third) with an evolutionary track to terabit rates built into the design of its backplane. The supplier's new gateway system incorporates a variety of large carrier-specific features with the basic H.323-standard voice gateway functions, including the option to translate circuit signals to either IP or ATM, a signaling gateway that allows SS7 (Signaling System 7) to be used in managing IP traffic and a feature server supporting direct input of advanced network services into the IP domain.
While Lucent plans to move to larger capacity gateways next year, the current capacity levels, at 30 DS-0s per gateway, provide room for a lot of growth, while supporting advanced features that will be part of those future systems as well, Simester says. "The beauty of the architecture is that, as carriers expand their markets, they can scale up by adding larger gateway servers and gatekeepers to handle more traffic within a zone and use the previously-installed smaller gatekeepers to serve gateways in lower traffic zones," he notes. (Gatekeepers are the servers that act as directory repositories for a cluster of gateways in the H.323 design, eliminating the need for each gateway to retain directory address information covering all the other gateways in a cluster.)
Such developments raise the obvious question of whether incumbent LECs and Sprint, moving to put ATM gateways in every business and home that subscribes to ADSL service, are imposing an unnecessary cost burden on themselves at the premises end just as IP edge and backbone solutions are obviating the need for QoS support from ATM. Sprint, BellSouth and even GTE seem adamant in their allegiance to the ATM-over-ADSL approach, which could be good news for competitors.
But don't count on it. If and when the day comes that QoS is readily available at carrier scales via IP, the new generation of edge switches they'll be deploying in the next year or two will be in place to support a move away from ATM at the premises. At that point, any remaining gap in the cost advantages enjoyed by new-generation IP-based competitors will become a thing of the past.