Welcome Guest. | Log In| Register | Membership Benefits


Network Hardware

Alternatives to high-speed networking abound, from Fast Ethernet to ATM
By Jan E. Johnson
Issue date: Oct. 23, 1995

he proliferation of client-server applications is causing companies to reassess the health of their corporate networks. Many companies are turning to high-speed networking alternatives to boost performance of desktop and server applications. But they're not just building corporate nets that suit current computing needs. Tec hnology managers are also laying the groundwork that will enable their companies to participate in what management gurus such as Peter Drucker call the networked economy--in which buyers and sellers worldwide are interconnected. Several options can help technology managers squeeze greater bandwidth from existing Ethernet and token-ring networks. They include FDDI (fiber distributed data interface), Fast Ethernet, 100VG-AnyLAN, and ATM. Almost all these choices involve transforming the network from a facility in which all nodes share the 10-Mbps pipe to a switched resource that can deliver a full 10-Mbps or even 100 Mbps of bandwidth to each individual node. In a sense, corporate networks are evolving to function more like the telephone network. This is a good thing, since PCs are increasingly used as communications devices that enable workgroups to collaborate across geographic distances.

Fast food purveyor McDonald's Corp. in Oak Brook, Ill., is a classic example of a company hungry for network bandwid th. "Two and a-half years ago, we had your basic token ring-Ethernet kludge," says Chuck Rush, global networking manager at McDonald's. "Networks were put in place independently by different departments and groups."

Slow Going
At the time, McDonald's had no strategy in place to manage LAN growth or departmental connectivity. Users ran standard desktop applications such as E-mail and Excel--certainly nothing exotic that would consume tremendous amounts of bandwidth, Rush recalls. Still, network performance was so sluggish that it could take as long as two hours to complete a simple print job.

As McDonald's grew, it began to implement client-server applications like Oracle databases and Lotus Notes. Network demands increased correspondingly. "At many companies, applications get developed in one place and the network gets managed in another," observes Rush. "And applications developers assume the bandwidth will always be there. Only in our case, it wasn't."

An FDDI backbone was McDonald' s first step to rev up network performance. The FDDI backbone delivers a dedicated 10 Mbps to each desktop. As a next step, Rush and his staff are prototyping an ATM (asynchronous transfer mode) backbone. To ensure up-to-date information on standards and products, Rush participates in the ATM Forum, a consortium of vendors and customers.

More than three-quarters of the large corporations polled by Forrester Research Inc., a market research company in Cambridge, Mass., acknowledge a shortage of network capacity. An avalanche of new LAN users and the expanded use of ordinary applications fuel this incessant drive for more bandwidth, says Paul Callahan, a network industry analyst with Forrester. Only 18% of those interviewed for the study ran real-time applications such as videoconferencing or online transaction processing (OLTP), both of which are notorious bandwidth guzzlers.

Many of these bandwidth issues are coming to light because centralized IS organizations are being asked to take charge of depar tmental networks and servers. Having a staff of network experts who can look at the entire corporate network allows the IS department to spot trends, witness recurring problems, and consider fixes that will benefit the entire organization, not just a single department.

But centralized network management isn't always enough. For example, though United Airlines has implemented centralized server farms that enable simplified administration and the sharing of expensive peripherals, when client workstations need to load applications that reside on these servers, precious network bandwidth is consumed, says Ken Cieszynski, senior staff engineer with United's architecture and technology group in Elk Grove, Ill.

United's network operates around the clock and is stretched to its limit. To help avoid a meltdown, the IS department has implemented an FDDI backbone as a stopgap measure, but it sees ATM as the long-term solution.

To vendors such as 3Com, Cisco Systems, Bay Networks, Hewlett-Packard, Digital Equ ipment, and IBM, this critical bandwidth shortage is an opportunity. These vendors are providing products in five key technology areas:

  • switched Ethernet (10BaseT)

  • hubbed Ethernet

  • switched Fast Ethernet (100BaseT)

  • 100VG-AnyLAN, and

  • ATM.

For technology managers, the variety of choices often makes it hard to determine the best alternative for a particular company's needs. There are more network choices than ever before, says Anthony Alles, ATM product manager for Cisco.

"Right now, there's a battle going on for what will be the next high-bandwidth solution for the desktop," says Mark Christensen, general manager of Intel's network products division in Hillsboro, Ore. "As a result, there's a lot of market thrashing and customer confusion out there."

To help ease that confusion, Forrester Research offers four rules:

  • Don't touch the desktop

  • Buy what you know

  • Don't walk away from a bargain

  • Avoid tech nology risk.

Desktops may be everywhere, but the majority are ISA bus machines sporting 10-Mbyte Ethernet or 14/16-Mbyte token-ring adapters, says Esmeralda Silva, a LAN industry analyst at International Data Corp., a research firm in Framingham, Mass. If the path to more bandwidth means sweeping changes to the desktop, the plan is not likely to be approved by IS. Drastic upgrades are too costly in terms of equipment and support resources to upgrade hundreds or thousands of PCs, Silva says.

In response to this marketplace reality, most Fast Ethernet, 100VG-AnyLAN, and low-end ATM switches are heavily populated with 10-Mbps Ethernet or 14/16-Mbps token-ring ports. These switches deliver bandwidth in a dedicated or shared mode, meaning the port's bandwidth can be delivered to single or multiple users who are connected to the switch through traditional Ethernet or token-ring hubs.

In this scenario, no changes are required on the desktop. Performance improvements come from dedicating bandwidth to a single desktop or from a reduction in the number of users on a shared network segment.

Vendors are delivering performance improvements to servers, too. Typically, a network switch supports one to four 100-Mbps ports, delivering dedicated bandwidth over a 100BaseT, 100VG, FDDI, or an ATM uplink module to servers, or routers. Performance improvements come from configuring servers in a star topology off of the switch. Valuable server cycles are released from the business of routing and segment management and can focus on applications tasks.

Greg Paquette, president of DataComm Products, a Maple Grove, Minn., reseller, expects 25% to 40% of his customers--medium and large corporations--to address bandwidth problems by moving to Fast Ethernet switches within the next couple of years.

Fast Ethernet (100BaseT) is an attractive upgrade option because network administrators already know how to support Ethernet. Fast Ethernet is backwards compatible with traditional Ethernet. Thus, network administrat ors can improve network performance without having to install new, unfamiliar technology. In essence, no retraining is required.

ATM is viewed as a new technology that will require lots of training as well as equipment upgrades throughout the network. Many large companies have ATM test beds, but few have production implementations.

Plenty Of Protocols
Hewlett Packard's 100VG technology is based in part on the Ethernet standard, but uses a different demand priority protocol and access method for getting bits on and off the wire. The protocol prevents any single node from hogging the network, while the access method is designed to avoid packet collisions, says Doug Small, HP marketing manager in Roseville, Calif.

The differences between standard Ethernet and 100VG AnyLAN are minimal, says Greg Scott, IS manager at Oregon State University's College of Business in Corvallis. 100BaseT lacks scalability, and ATM isn't ready for prime time, Scott says. But 100VG is a solid, scalable solution for supporting the bandwidth demands of Microsoft Windows 95 and potential video-based applications.

In addition, Fast Ethernet appears to top out at 100 Mbps, while the 100VG standard will easily scale to 400 Mbits and, theoretically, into the gigabit range.

The State of California's Employment Training Panel opted against 100VG because of limited vendor support, says Brian Cundiff, associate IS analyst with the agency. More vendors are going with Fast Ethernet because if offers backwards compatibility with standard Ethernet. Cundiff installed Fast Ethernet switches, providing dedicated 100-Mbps pipes to servers, while supporting most users with dedicated 10-Mbps links. A small group of power users share a 100-Mbps pipe.

Another drawback to 100VG is its incompatibility with existing network protocol analyzers. An existing LAN analyzer can't be hooked into a 100VG network without some changes to the product, explains Chip Mesec, product marketing manager at Network General in Menlo Park, Calif. Network General has no plans to build a line of 100VG analyzers. "We're building Fast Ethernet and ATM Sniffers," Mesec says, "but we haven't seen enough movement in the 100VG marketplace to warrant our entry."

Big improvements in switch management lie ahead, if the Virtual LAN (VLAN) management standard pans out. VLAN is a management standard that will work with all fast switching technologies except ATM. With VLAN, smaller IS staffs could more easily manage and administer additions, moves, and changes on large switched networks, predicts Jayshree Ullal, director of marketing at Cisco. VLAN is designed to give IS managers the tools to group physical nodes on a switched network into logical domains. For instance, employees in the finance, marketing, and engineering, and their corresponding PCs, servers, and printers, make up three primary domains. Finance employees around the world would be administered as a single group or domain and assigned a common set of network resources.

Suppose a special proje ct requires one person from each of those groups to work together for six months. These users would be defined as a single group, administered as a unit, and assigned a specific set of network resources. The task force members would retain their primary group privileges and gain a new set of network privileges associated with the task force.

The price of Fast Ethernet is "dropping through the clouds," says Network General's Mesec and others. Intel's 10BaseT/100BaseT PCI card now lists for about $150, while HP's 100VG card lists for about $200, placing both in the10-Mbps, PCI card range. The price of FDDI adapters, which list for about $1,000, keep FDDI from being considered a long-term viable alternative for boosting LAN bandwidth.

FDDI is rapidly becoming a legacy technology at many large companies, according to Forrester Research's Callahan. "With 100BaseT cards priced in the $100-to-$200 range, it's no big deal to go to 100 Mbps," says Cundiff. Many IS managers are planning to "future-proof" their incoming Pentium PCs with LAN adapter cards that support both 10 Mbps and 100 Mbps. Within two years, Cundiff expects to have 100 Mbps cards in all desktops. "Then it will be a matter of replacing the hubs," he says.

Bandwidth Galore
ATM technology is still in its infancy, and at this early stage of development, it violates all four of the LAN administrators' golden rules: It requires new adapters for desktops, servers, and hubs, as well as a complete change in network paradigm. In addition, it's still quite expensive and carries serious risks.

As a result, corporations will continue to use and upgrade frame-based options like Fast Ethernet. ATM will be revived only if all other frame switching alternatives fall short on delivering priority and reserved bandwidth, says Forrester's Callahan. These hurdles will continue to relegate ATM to a minor role in backbone networks for the next 10 years, he adds.

Still, ATM is gaining hold in market niches, such as print production, video, and TV broadcast applications, medical imaging, financial transactions, and seismic analysis. These are areas in which companies need to move large amounts of data--and where the need for speed far outweighs the risks.

Going with a single-vendor ATM solution is one tactic early adopters employ to reduce risk. "When you're at the bleeding edge and standards are not fully set, you need a single-vendor solution," advises Scott Knipfing, VP of the finance desk for Donaldson, Lufkin & Jenrette, a New York brokerage.

The brokerage chose Fore Systems Inc. in Warrendale, Pa., for its ATM system, including switches, ATM cards and drivers, and network interface cards. Four ATM switches support 35 desktops and a file server. All desktops have Windows NT 3.5 and talk to a Novell NetWare file server using 100-Mbps or 155-Mbps ATM cards.

The file server gets a dedicated 155-Mbps link off the switch, says Knipfing. His hot tech tip: "Give the file server gobs of memory." He configured his Pentium server with 280 Mbytes of RAM. "When we tested our ATM configuration with a real-world application, the re-indexing of our trading file took just five minutes, compared with 15 minutes over the old network," says Knipfing. The old configuration was an FDDI backbone with 100 Mbps to the desktop and two LAN segments linked to the file server.

Time wasn't all that the brokerage saved. ATM also enabled the Finance Desk to reduce daylight overdraft fees paid to the Federal Reserve. The Fed charges brokerages a fee for every minute that the firm possesses U.S. Treasury bills, notes, and bonds. The faster the brokerage can close a trade and get the bonds out the door, the less it pays. Knipfing figures the money saved from reduced Fed fees paid for the ATM equipment within 10 months. Donaldson, Lufkin & Jenrette is on the verge of a corporatewide rollout of its ATM network.

Phoenix Newspapers Inc. is also implementing ATM. Unlike Donaldson, Lufkin & Jenrette, which implemented it to traders' desktops, the newspaper publisher is putting ATM into its backbone network only.

"Virtually no legacy equipment will have to be supported," says Allen Hsu, network architect and lead project manager. The new network will support a mixed bag of protocols, including TCP/IP, IPX, AppleTalk, DECnet, press equipment, and a Honeywell mainframe.

The reason for using ATM in the newspaper's computing environment is simple: The business is becoming more image intensive, says Hsu. He points to ongoing experiments in unique presentations of news over the Internet, and the prospect of video over the Internet in the future.

In Phoenix Newspapers' initial installation, desktops will be linked to the Cisco ATM switch via 10 Mbps and 100 Mbps Ethernet ports, supporting a mix of dedicated and shared bandwidth. "ATM is wonderful as a backbone," says Hsu. "But there's no reason to pull it to the desktop. It's too expensive, and users don't need it."

Standards Evolving
By year's end, two key standards are expected to emerge from the ATM Forum. One addresses traffic management; the other, routing. The solidification of these two standards will cause substantial changes in product design, resulting in a new generation of ATM switches in 1996, predicts Cisco's Alles.

Several additional ATM standards on the horizon will address limitations in ATM technology. For instance, today's ATM networks, if not managed closely, can easily become congested and collapse, says Alles. The traffic management protocol, Available Bit Rate (ABR), addresses this problem by defining a congestion control feedback mechanism between the switch and attached end points.

The ATM Private Network-to-Network Interface (PNNI), defines how to route a connection through a series of ATM switches. "The problem of routing an ATM request from source to destination is similar to routing a packet through a local area network," Alles explains.

Although the PNNI protocol builds on many of the same concepts used in routing today, it's probably the most complex protocol ever developed. "I don't expect to see PNNI commonly deployed and fully interoperable for at least a couple of years," Alles says.

Address resolution between TCP/IP and ATM, for instance, is another complex task that concerns network administrators. The ATM network, unlike a traditional LAN, has to know the data's destination. The industry's first attempt at solving this problem was the LAN Emulation standard. Released early this year, the standard has shown up in various vendors' ATM equipment.

Multiprotocol Over ATM is the next step toward effective address resolution and LAN interoperability. It should be out in the second half of 1996.

It's already difficult to tell whether a PC application is running locally or across the corporate net. With advances in high-speed networking, the seamlessness of corporate computing will increase. With unbridled network speed, the promise of video teleconferencing and global information access will become a reality. Companies will achieve new ways of doing business that we've only begun to imagine.

Click here for a related story on Amoco's ATM project

Comments on this story?

InformationWeek http://techweb.cmp.com/iwk