Is ATM ready for prime time -- yet?
Sun's 622-Mbps ATM adapter has a technology edge, but how does it play in the real world?
Even though it's costly, a one-of-a-kind 622 megabit-per-second (Mbps) asynchronous transfer mode (ATM) adapter from Sun outpaces the market and delivers high bandwidth for at least one user, Texas A&M's department of computer sciences. (1,800 words)
Hey, you! Yes, you with the souped up SPARC servers and a LAN so bogged down with traffic that the little old lady from Pasadena would feel safe on it.
Have I got a deal for you: Virtually unlimited bandwidth for your server, guaranteed on-time delivery of traffic generated by mission-critical client-server, video and imaging applications, and scalability beyond your imagination.
And it's all yours by simply installing one of Sun Microsystems' relatively new SunATM 622 asynchronous transfer mode (ATM) adapters, which delivers 622-megabit-per-second (Mbps) data rates over multi-mode, fiber-optic cable, in your SBus-based SPARCserver 1000 computers. Then watch out: You'll have more than enough horsepower to "overload" your client computers, says Texas A&M Senior Lecturer and Departmental Computing Services Director Willis Marti, one of the chosen few who has actually put the SunATM 622 card through its paces.
Watch out! Here's the fine print
Oh, did I tell you you'll have to totally redesign your LAN infrastructure, not only upgrading your existing copper cable to multi-mode fiber but deal with switched and permanent virtual circuits? And that you'll have to fork over $4,995 for each adapter, plus another $7,995 for a switch module to integrate it into your LAN (and that doesn't include the cost of the switch itself, which can run another $50,000 to $100,000)?
Did I forget to mention that you may not be able to integrate your ATM-based network into your legacy LAN? Or, if you're willing to wait another year or so, you'll be able to buy gigabit Ethernet products offering even better performance at half the price of ATM without forcing you to totally overhaul your network infrastructure?
By all accounts, vendors have experienced nothing but trouble trying to sell ATM at any speed, whether it's the lower-priced, 25.6-Mbps version targeted at desktop connections and the 155-Mbps LAN backbone rate (also known as OC-3), let alone the expensive, and slowly emerging, 622-Mbps (OC-12) variation. In fact, two recent market research reports indicate ATM's growth will remain stunted until at least the year 2000.
In its "User Plans for High-Performance LANs, 1997," report, Infonetics Research of San Jose, CA, discovered that only 15 percent of 107 corporations with switched 10 Mbps or faster networks would use 155-Mbps ATM in backbone installations; only 2 percent said they'd connect servers with 155-Mbps ATM. Only one company said it planned to deploy 622-Mbps ATM by January of this year in any application, says the study's author, J'Amy Napolitan, Infonetics' chief operating office and senior analyst.
Dataquest numbers echo those from Infonetics. "We expect ATM to get about 2 percent of the worldwide NIC [network interface card] market by 1999," says Veronica Guerrero, Dataquest's NIC industry analyst.
Just what's holding back ATM?
The reasons for ATM's struggle are many, and varied. First, and most obviously, most MIS departments have been put off by the costs associated with deploying ATM. As noted, Sun's 622ATM NIC sells for $4,995, while Fore Systems Inc.'s (Warrendale, PA) corresponding switch module, required to give the company's ForeRunner switch OC-12 capabilities, costs $7,995.
Even the 155-Mbps ATM NICs available today are pricey: Adaptec Inc.'s (Milpitas, CA) offering sells for $495, while Fore and Madge Networks (San Jose) offer 155-Mbps cards at $995. In addition, Newbridge Networks Inc. and Interphase Inc. (Dallas, TX) have announced a 155-Mbps ATM adapter, one that supports the ATM Forum's new multiprotocol over ATM (MPOA) standard, for $1,495.
Contrast that with 100-Mbps Fast Ethernet NICs from 3Com Corp. (Santa Clara, CA) that cost from $94 to $135 in quantities of 100, and Intel Corp., which sell for $84 to $129 in the same quantity. At the switch end of the connection, Fast Ethernet runs as low as $65 per port, with ATM still running in the $1,000 to $1,500-per port range.
But where is the "killer app"?
Foremost among the other realities that have blunted ATM's deployment is the lack of a true "killer app" that would make the technology a "must have." As market analyst Tom Nolle, principal of the CIMI Corp. (Voorhes, NJ) noted in his January 1997 Netwatcher newsletter, "The glamorous applications like multimedia have largely disappeared from ATM booths (at trade shows). This shift away from new applications both reduces the dollars available to justify ATM (by reducing the business benefit pool that multimedia would have augmented) and changes the feature emphasis.
"The benefit reduction is a key issue for ATM, because only vestigial justification for the exaggerated forecasts for ATM deployment was the expectation that a new source of benefits would justify the new costs," Nolle continued. "Without those, ATM purchases would have to fall short."
"Even at 155 Mbps, ATM is very rare," Nolle said in a recent phone interview. "In 1996, 25-Mbps ATM outsold all other forms of ATM altogether, and we don't predict any significant 622-Mbps sales until after the year 2000. It's a `hobby' market -- if someone has the budget, then they buy it. That's what we call the federal and university R&D [research and development] market -- someone gets a grant, and they buy a trinket, but there's no business market for [622-Mbps ATM]."
Competition from Fast Ethernet and the impending gigabit Ethernet also has played a key role in ATM's slow growth. "Gigabit Ethernet does offer an attractive alternative and, at the present time, a lower-cost alternative," agrees John Armstrong, principal networking analyst for market research firm Dataquest.
The Infonetics study of 107 high-performance LAN users backed this claim. Sixty-three percent of the respondents said they'd use gigabit Ethernet instead of ATM "in general," while 52 percent thought gigabit Ethernet would delay ATM's deployment.
Not that gigabit Ethernet products are all that inexpensive now, either. Alteon Networks Inc.'s AceSwitch gigabit Ethernet switch, for instance, lists for $8,995 for eight 10/100-Mbps Ethernet ports, two one-gigabit-per-second ports, and an expansion slot for a gigabit Ethernet expansion module. Its first gigabit Ethernet NIC is priced at $1,495. Another vendor, GigaLabs (Sunnyvale, CA), offers a two-port gigabit Ethernet switch for $11,000; it is also shipping a $3,495 gigabit Ethernet NIC for PCI and SBus systems.
What do users say?
Despite all that, ATM still offers some compelling features, says Texas A&M's Marti, one of the initial beta users of the SunATM 622 adapter. Only ATM could deliver the throughput required by users in the school's computer science department, which runs a variety of "complex" applications, such as VLSI (very large system integration) simulations and testing, network simulations, and robotics over a distributed environment in its College Station, TX facilities.
Prior to its participation in Sun's ATM622 beta test, the Texas A&M computer science group's departmental server, a SPARC1000, was connected via Ethernet to multiple Ethernet subnetworks as well as the campus backbone network (via a Fore Systems' router). That solution became unacceptable for two reasons: The server couldn't scale beyond supporting four Ethernet segments directly, and end-user, server-to-client performance declined precipitously for the workstations on the distant side of the router, says Marti.
Because the students and staff of the department are extremely "mobile, every workstation (in the department) has to be able to support all of the applications and environments" in use within the department, says Marti, who teaches a senior-level course in networks and distributed processing. And the complex nature of those applications -- programming, engineering and the like -- meant each workstation needed more bandwidth than a 10Mbps-Ethernet connection could handle.
"The servers had more `gas', but the clients were getting extremely poor response," says Marti. "So we decided to get a bigger `pipe'." That "pipe" is a 622-Mbps ATM connection between the departmental SPARCserver 1000 server and an ATM switch from Fore Systems. A router connects the department's workstations to the ATM switch via a variety of subnetworks.
"We tried FDDI (the 100-Mbps Fiber Distributed Data Interface), which helped a lot," but didn't solve the department's bandwidth problems, says Marti. He also considered a 100-Mbps Fast Ethernet, but believes that technology "effectively is a dead end."
During the initial beta testing, when Texas A&M also used several SunATM 155-Mbps ATM cards, Marti says he was able to get "about 100-Mbps throughput to our clients. With the 622-Mbps card, as we upped the load (from the server), we found out we could `top off' the clients." Throughput was linear (in relation to server output) until we overloaded the workstation, and that was what we were looking for.
"One of the reasons why we participated in the (SunATM 622) beta program rather than just buying sight-unseen is we wanted to see how well they work, and they work real well," explains Marti. "After returning the beta board to Sun, Texas A&M purchased two of the SunATM 622 NICs and was planning to install them in a pair of compute servers. Marti says he plans to purchase more of the SunATM 622 cards as the department installs more of the fiber-optic cable required to connect the cards to the Fore switch.
"They're not cheap," Marti admits. "Would I buy them based purely on costs?" he asks. "I might not do that, but we're not totally cost-driven in the computer science department, either."
Competition in the
The costs of 622-Mbps ATM NICs could begin to fall within a year or so, however, as more and more vendors enter this market niche. A quick SunWorld survey of ATM NIC vendors found that at least three of them, Interphase, Adaptec, and Fore Systems, planned to market 622 Mbps cards "in the future," with no firm shipment dates.
Of those, Adaptec said it probably would not offer an SBus product. "The SBus peripheral market is very small," explains David Mendenhall, ATM marketing manager at Adaptec. "Less than 10 percent of the peripheral market goes to (Sun) third parties, so it would make more sense for us to move to the PCI bus."
Interphase's Les Zsohar, the company's principal engineer, says Interphase will ship a PCI bus 622-Mpbs ATM NIC "late in the third quarter or early in the fourth quarter," and that an SBus version is "an option."
Laurie Sims, Fore Systems' ATM NIC product manager, says, "The SPARC market has one of our largest bases of installed OC-3 (155 Mbps) NICs, and right now, the only application for OC-12 (622 Mbps) would be in Sun servers," so it's natural to expect Fore Systems to market a 622-Mbps ATM adapter for the SBus.
Here's another bit of "science fiction" from Fore Systems: Seth Redmore, the company's ATM switch product manager, says that "we've even been asked for workstation-based OC-48 (2.5 Gbps) capabilities, and we're looking to have that sometime over the next year."
For now, however, 622-Mbps ATM represents the top-of-the-line network buy -- even if it's not all that great a "deal" yet.
About the author
Jim Carr is a Saratoga, CA, freelance business and technology writer who has covered the networking industry for more than 10 years. Reach Rick at firstname.lastname@example.org.
If you have technical problems with this magazine, contact email@example.com