s. And thanks to the rise of Gigabit Ethernet, there will be no shortage of bandwidth among servers.
That leaves the traditional argument for using ATM: to collapse many different networks -- voice, video, and data -- onto a single backbone. But ATM doesn't
look like the only way to do even that anymore. Instead of ending up as most things to most people, ATM will turn out to be some things to some people -- particularly phone companies that have already climbed the ATM learning curve. The increasing speeds at which frame relay runs, along with the promise of new IP services such as guaranteed bandwidth and voice over frame, are challenging ATM's assumed dominance as a public WAN service. Furthermore, frame relay is based on IP addresses, whereas ATM's addressing scheme is based on ISDN phone numbers. With IP-based services at the local exchange carrier and offered by many Internet service providers (ISPs), it will be difficult to "dial" others using an ISDN-based system they aren't subscribed to. The growth of frame relay, coupled with new technologies to speed packet services and counter congestion, promises to preserve familiar IP addresses and routing protocols (such as OSPF) instead of forcing a migration to whole new schemes.
If you haven't yet impl
emented the technology, and especially if you're not a phone company, the bottom line is this: Get ready for an explosion of reasons not to incorporate ATM into your customer premises equipment (CPE). The age of IP dial tone is at hand.
ATM and Frame Relay
If you thought ATM was the only way to get a high-speed WAN connection, think again. Frame relay is breaking through its T1 (1.544 Mbps) and T3 (45 Mbps) speed barriers. Ascend Communications is running frame relay at 155 Mbps in its lab today, according to Dick Kachelmeyer, the company's director of product marketing.
Thanks to the Internet Engineering Task Force (IETF) and the Frame Relay Forum, frame relay is also gaining some decidedly ATM-like features, including voice, guaranteed bandwidth, and flow control management. One of the most important of these is FRF.11, a standard for voice over frame relay, which the Frame Relay Forum ratified in May.
Also, by the time you read this, the Frame Relay Forum should have approved a f
ragmentation implementation agreement that will outline how to break frame relay frames into smaller frames. This agreement will give frame relay even more ATM-like capabilities, such as quality of service (QoS) levels, which could be mapped to equivalents in ATM hardware through interworking, says Larry Greenstein, vice president of technology for the Frame Relay Forum.
Also this year, the Forum hopes to finalize service-level agreements (SLAs). These would let carriers describe their services to users, then let those users measure the service they're getting to determine if they're getting what they pay for. While frame relay's existing committed information rate (CIR) is a way of determining the minimum rate at which frames get sent over a connection, SLAs could let customers or carriers specify the number of frames that could be discarded over a given time period, and provide customers with financial refunds if that number is exceeded. Despite concerns that the new standards would require frame rela
y hardware to be upgraded, manufacturers such as Ascend Communications say the new features will require only a software upgrade. In fact, Ascend plans to release its version ahead of the standard, then upgrade to meet the standard when it's completed.
So, think it's time to jettison ATM for frame relay? Not quite. For starters, OC3-speed (155-Mbps) frame relay has a long way to go: Ascend has to announce and deliver products before service providers can roll out the technology. Moreover, the Frame Relay Forum group isn't working on any standard frame relay speeds beyond T3.
Second, voice over frame relay isn't ready for widespread use over public networks yet. "If the network experiences peak traffic and congestion, voice doesn't perform so well," says Heidi Brandte, senior product marketing manager at Ascend Communications. It's mostly useful for intracompany voice traffic today, she admits. Carriers such as Bell Atlantic hope to announce voice over frame relay services by the end of this year.
"Today, if you allow a large LAN traffic frame to go in between voice frames, it will obviously affect quality," according to John Rolfe, senior product manager for frame relay at Ascend. Fragmentation will help frame relay reduce latency and deliver advanced services -- even video, Rolfe says.
When you get down to it, voice is just plain tricky. Even ATM still has some problems dealing with voice. While some proprietary solutions, such as Fore Systems' ForeRunner VoicePlus network module, shipped earlier this year, the ATM Forum standard to provide plain old telephone service (POTS) to PC desktops still lacked a number of features at press time, including the ability for a user to hear a busy tone. The enhancements needed were headed for final ballot by early August. Even so, the proposed standard won't work with anything other than constant bit rate (CBR) ATM, which provides data at a guaranteed rate with rigorous latency control. "There are some difficult timing issues that need to be worked ou
t" to get voice to run over ATM's more cost-effective variable bit rate (VBR), says George Dobrowski, president of the ATM Forum.
Ultimately, packet-based services are less than ideal for handling high volumes of private branch exchange (PBX) phone calls. If WAN traffic is to include PBX-to- PBX traffic, it has to carry clocking information, the output of old time division multiplexers, across the network. "There's still a huge legacy phone system," says Ascend's Rolfe. It's a phone system that doesn't tolerate the kind of jitter, or variations in latency, common on packet networks. Unless thousands of legacy PBXes suddenly add buffering, it'll be packet networks that have to adjust.
ATM, IP, and QoS
How will they adjust? Try IP. The future of WANs could hinge on whether anyone can figure out how to provide priority service for critical traffic. The IP camp has Resource Reservation Protocol (RSVP), an imperfect scheme at best. RSVP relies on network devices, such as routers, to make a bes
t-effort attempt to deliver isochronous traffic, such as video. It may, however, initially be best at simply prioritizing non-time-sensitive packets that can still live with some latency.
ATM, of course, already specifies QoS classes that can guarantee end-to-end latency. But at a price: Once an ATM switch reaches its capacity of virtual circuits, the switch refuses additional connections, and routing must again commence to carry excess traffic around the congestion.
The debate about how to end congestion in switches and routers rages. IP fans believe that new technology, such as MMC Networks' Xstream chip set (
see the figure
), implemented in Cisco's new LightStream 1010 router, lets IP as well as ATM switches give isochronous traffic priority.
ATM proponents insist that it has to be done with ATM. "The average packet traveling across the Net takes 16 hops," says Dave Nelsen, senior marketing director at Fore Systems, a leading provider of ATM switches. "About half of t
hose occur on the backbone. When you put in ATM as a replacement backbone and push the routers to the edge of the ATM core, traffic can move directly from the access router to the egress router with no router hops in between."
As a way of eliminating the need for routers, telecommunications companies are also rushing to deliver switched virtual circuit (SVC) service for ATM. SVCs will offer ATM customers more flexible usage-based billing, and they are more affordable for lower-usage customers than permanent virtual circuits (PVCs), according to Nick Nechita, senior architect of broadband technologies and service for the New Brunswick Telephone Company (Saint John, New Brunswick). In the U.S., AT&T recently became the first interexchange carrier (IXC) to offer ATM SVC as a public service. IXCs are also widening their ATM pipes, from OC3 and OC12 today to OC48 within 12 months. This is one area where frame relay is lagging.
Both approaches have their merits. As long as the Internet keeps growing
, applications that need QoS will still experience brownouts and blackouts. In fact, there's even an effort to bypass the debate not by switching all traffic but by building faster routers. Far better, critics say, to maintain the existing democratic routing hierarchy, which gracefully degrades service but does not deny it.
So, would you rather have affordable videoconferencing service with variable quality, pay for a service that could have busy signals, or just stick with pricey point-to-point systems? You may be asking that question whether you go with ATM or stick with IP on your WAN.
Rough Seas
Even ATM's strongest proponents now concede that public WANs, including connections to ISPs, will be a mixture of frame relay and ATM. Phone companies' ATM support on their T1 lines is increasing dramatically, but ATM will still be playing catch-up to frame relay, which is already offered in practically every market.
But frame relay's lack of SVCs impacts the ability of providers to char
ge sensibly for it, and for customers to know what they're paying for. "It's very hard to count IP packets," says David Dorman, chairman, president, and CEO of Pacific Bell. "It's easy to count how long a circuit has been open and who opened it." The phone companies continue to push hard for this to become a part of IP services, so Internet access can be metered instead of flat-rate. If current trends continue, by 1999 more than half of Pacific Bell's traffic will be data, not voice, Dorman says.
Despite technological challenges and slower-than-hoped acceptance, ATM represents a healthy business. Frame relay growth has slowed only to double digits, while ATM remains in triple-digit territory, according to both the ATM Forum and the Frame Relay Forum. When you add up equipment and services, both are billion-dollar-a-year industries.
Where ATM makes sense today is at the core of some very large networks. Phone companies, for example, remain bullish on ATM pushing its way to the very edge of the Int
ernet. "ATM has traffic management capabilities, segregation, and prioritization of traffic," says Andy Schmidt, product manager for Ameritech Data Services. "It's very difficult to get that done with IP alone." Sixty percent of today's Internet traffic, including frame relay, is carried across backbones in ATM cells.
But all the value-added services ATM promises -- voice, video, variable bit rate transmission -- have been late in coming. The reason: ISPs are doing all they can just to keep up with demand for existing services. Bursty, Web-based Internet traffic doubles every three or four months, according to Alan Taffel, vice president of marketing at UUNet Technologies.
Where to Find
Ascend Communications
Alameda, CA
Phone: 510-769-6001
Internet:
http://www.ascend.com
ATM's Shrinking Role
