5/4/98 InfoWorld article. Fairly positive on ATM vs. GE
Fortify your backbone. Meeting the need for network speed
infoworld.com
Related charts:
Switching to ATM (chart)
Network factors driving high-performance LAN purchases (chart)
By Stephen Lawson
As thousands of CIOs and LAN administrators descend on Las Vegas this
week for NetWorld+Interop, many will be weighted down with infrastructure
woes even before they pick up their complimentary tote bags.
Some are confident that the hubs, routers, and switches in their networks
today will accommodate all the new applications that users will throw at them
for the next few years. But for others, the time has come to make the big
cut-over to a backbone technology that can take them into the next
millennium.
Before long, those bags will be filled with brochures, mouse pads, and
gimmicky pens from vendors touting newly hatched technologies to solve
those problems. How managers choose from that grab bag, and how they
implement the migration, will make all the difference in how their networks
hold up in the coming years. The key, in most cases, will be fitting the right
technologies into the right roles in the network and making them work in
harmony.
Observers say that a number of factors are combining to make traditional
routed networks obsolete in many cases, driving many users in search of
newer switching-based gear to fortify their connectivity infrastructure.
"The software-based routers installed in a lot of backbones are coming under
a lot of pressure from things like the Internet, intranets, and growing
applications requirements," according to Esmeralda Silva, an analyst at
International Data Corp., in Framingham, Mass.
The advent of the Web has opened up a world of resources -- and
distractions -- that can send end-users outside the LAN in droves, taxing
routers that were installed back when most traffic was confined to the local
area.
Intranets are creating the same effect inside the campus, with users reaching
outside their local network segments to servers in other departments or in
central server farms. Intranet applications that run on Web browsers also
increasingly take advantage of graphical content, and emerging applications
bring voice and video into the mix.
"This type of stuff will be on the network by end of the year whether the
network is ready for it or not," says Melinda LeBaron, an analyst at the
Gartner Group, in Stamford, Conn.
And even the good news about networking can be bad news. As switches
come down in price to compete with shared-media hubs, in some cases
providing a dedicated 10Mbps or 100Mbps pipe out of each end-station,
those streams of data converge into a rushing river at the center of the
network. LAN backbones that provide 100Mbps of throughput under the
best of circumstances, such as Fast Ethernet and FDDI, and routers that may
process only 250,000 packets per second, are running out of steam,
according to analysts and users.
NEW FANGLED GEAR. Out on the floor, show-goers shopping for
backbone gear will find several crops of technologies that are blooming now
for the first time. They will be able to place orders for Gigabit Ethernet
equipment from a wide variety of suppliers, including some of the top
vendors. Routing switches, now in the works at almost every hardware
vendor and working in the booths of many, combine the speed of
hardware-based packet switching with some routing intelligence. And ways
to guarantee quality-of-service (QOS) via packet-based networks, though
still largely on the drawing board, are being designed into many new devices.
Emerging technologies for ATM, including Private Network-to-Network
Interface and Multiprotocol over ATM, are starting to provide greater
routing intelligence and integration with packet-based networks. And some
major suppliers of Token Ring, including IBM and Olicom, will demonstrate
a version of that technology boosted to 100Mbps.
All these options will make it difficult for network planners to design a
backbone for the future.
Observers say the choice between Gigabit Ethernet and routing switches on
the one hand, and ATM on the other, may not be black-and-white. To meet
users' needs from the desktop across the LAN or WAN, some organizations
may have to implement both. But there are benefits and drawbacks to each
that can help determine where they can best be used.
PACKET-BASED NETWORK UPGRADE. Ever since the advent of
LAN switches, enterprises building or expanding packet-based networks
have been faced with a dilemma. Switching can boost the forwarding speed
of a network, but it does not afford the control and security that come with
routing. However, traditional routers, which can break up the network into
manageable and secure segments, incur overhead because they use software
to examine and direct every packet.
Routing switches, or Layer 3 switches, are designed to remove this dilemma
by routing the most common types of packets at the speed of switching.
Add to this prospect the additional performance of Gigabit Ethernet, and
packet LAN technology is set to leap ahead in speed just when it's needed.
"If the majority of your traffic is data, going the Layer 3 route may be a pretty
reasonable thing to do, depending on the size of your company," Silva says.
The University of North Carolina is preparing to test Cabletron's
SmartSwitch Router, a gigabit-speed routing switch chassis developed by
start-up Yago Systems. If the SmartSwitch Router works as advertised,
according to an IS administrator said, such devices eventually could replace
Cisco 7500 routers in several places on the campus network.
"Being able to get the routing processing off of software and into hardware
would help," says Jim Gogan, director of networks and communication at the
University of North Carolina, in Chapel Hill, N.C.
Although the network is not sagging yet, Gogan is planning ahead.
"If we don't look at another alternative in the next few months, we'll have a
problem six to eight months out," Gogan says. "I'd rather be proactive and
make changes like that gradually rather than reacting."
FAMILIAR PACKETS. A key advantage of supercharging a packet
network with Gigabit Ethernet and hardware-based IP and IPX routing,
according to users and analysts, is that a staff used to working on Ethernet
won't need to be retrained for an ATM infrastructure. In addition, the same
packets can originate at the desktop and be carried across the LAN
backbone without conversion to cells along the way, which can cause delays.
And administrators won't have to map Ethernet's addressing scheme to
ATM's through LAN Emulation, a job that can lead to a lot of
head-scratching.
The limitations of investing in a packet-based upgrade come into play if an
enterprise plans to adopt applications that will send delay-sensitive traffic,
such as Microsoft's NetMeeting or other videoconferencing tools, call-center
software, or computer telephony. Makers of switches and routers have
announced plans to leverage a wide variety of technologies to assign and
enforce priority and latency on packet networks.
Some observers warn that if you want to be able to prioritize traffic on the
basis of policies, a packet-based network will be more difficult to manage
than one based on ATM.
"In some sense, the burden is on the network manager to understand what
the policies will do," LeBaron says. "You have more options to screw up the
network than you've ever had before."
HEADROOM AND ATM. Most users who have chosen the path of
ATM cite the technology's built-in QOS features, its maturity, and a clearer
path to higher speeds in the future.
Observers say standards for ATM, long in development, finally have
stabilized. And recently introduced specifications promise to simplify
configuration and management of ATM backbones.
The increasingly dominant role of an ATM in carrier networks also makes it a
good choice for large, far-flung enterprises that need high-speed connections
to branch facilities, analysts said.
The NASA Classroom of the Future, which uses video-based training
extensively over a campus network of approximately 2,000 users at
Wheeling Jesuit University, in Wheeling, W.Va., began adopting ATM on its
backbone in 1994. The built-in QOS features of ATM give video traffic the
steady bandwidth and latency it requires across the core of the network. But
cost constraints prevented deployment of ATM all the way to the desktop,
so switched 10Mbps Ethernet suffices.
"From the desktop, we're still doing it the old way, which is throwing
bandwidth at the application," says Nitin Naik, executive director of the
NASA Classroom of the Future program at Wheeling Jesuit University.
ATM also has the advantage that it was built to scale up to higher bandwidth.
Because it was designed for carrier networks and uses Synchronous Optical
Network technology, ATM LAN solutions have already increased from
OC-3 (155Mbps) to OC-12 (622Mbps) and are set to move to OC-48
(2.4Gbps), which is currently used by some carriers.
Ethernet, by contrast, has no path toward higher bandwidth save trunked
gigabit links and a fledgling proposal for a 10Gbps specification. This
difference could affect future infrastructure costs.
"I believe that it's going to be less disruptive to increase the overall capacity
of an ATM backbone than Ethernet backbones," says Mary Petrosky, an
analyst at the Burton Group, in San Mateo, Calif.
"The fact that ATM is scaling at a more reasonable pace makes it more
reasonable to take an existing ATM box and add these faster interfaces,"
Petrosky says.
Jim Atwood, a network analyst at Egelston Children's Health Care System,
in Atlanta, says his organization implemented an ATM backbone to provide
for videoconferencing via both the LAN and the WAN.
For example, doctors will be able to consult with patients and view X-ray
and other records at remote facilities, without leaving the medical center
where they are based.
HOW TO PROCEED. Just as the choice of a backbone involves more
than the technical merits of a particular approach, observers and users say,
planning for and carrying out an upgrade is more than a matter of drawing a
topology and writing a purchase order.
Planning a backbone upgrade means thinking beyond applying a certain
technology to a particular need, LeBaron says.
"When you're in reaction mode, you don't always get the best performance
out of the network," LeBaron says. "The network gets Frankensteined."
To start with, analysts and users say that those in charge of technology in an
enterprise should bring others into the decision-making process. Users know
what the network can do now and what they want it to do in the future.
"For any upgrade, you need to know what level of activity is going on in your
network," Naik says. "[Then] try to find out from the user side what sort of
applications they perceive happening in the future of the organization."
Fortunately, such an approach is easier to sell than ever before, according to
LeBaron. Many executives now recognize that strong links among employees
and to customers and partners outside are crucial to a business. Because top
management now views the network as a key asset, or even a profit center,
organizations are planning for it as they would plan factories or investment in
growing markets.
LeBaron recommends that companies map out all their technology needs,
including for servers, client systems, and applications, before creating a
strategy. Teams from across the organization should decide what they need
to make the company successful, she says.
When bids are requested and network, server, and application vendors
respond, users should still keep hold of the reins.
"Bring vendors in together and find out how they're going to make the
network work with applications and servers," LeBaron says.
Stretching out deployment over a period of time can have benefits, too, as
chip development inevitably makes hardware faster and cheaper.
"Moore's Law is there, so please don't forget it," says a network manager for
a large engineering company, who is planning the infrastructure for a campus
that will serve thousands of users throughout many years.
If the recent past holds any lesson, it is that any projected need for
bandwidth -- and any technology's imagined capacity to fill that demand --
may soon be outstripped by reality.
Senior Writer Stephen Lawson covers networking. He can be reached
via e-mail at stephen_lawson@infoworld.com.
For rundown of the most recent feature stories on InfoWorld Electric
see Features at a glance.
Questions or comments? Send an e-mail to our Editors.
Go to the Week's Top News Stories
Please direct your comments to InfoWorld Boston Bureau Chief, Ted Smalley Bowen
Copyright c 1998 InfoWorld Media Group Inc.
InfoWorld Electric is a member of IDG.net
| SiteMap | Search | PageOne | Conferences | Reader/Ad Services |
| Enterprise Careers | Opinions | Test Center | Features |
| Forums | Interviews | InfoWorld Print | InfoQuote | |
