FOLLOW THE IP SWITCH ROAD [Nice AT&T/ASND references]
internettelephony.com
ANNE BILODEAU ZIEGER, Cover Story, July 20, 1998
The definition of Internet protocol switching has shifted a few times since its March
1996 inception. The term once referred to something quite specific: a software
enhancement to asynchronous transfer mode switches that enabled them to move IP
packets more efficiently. Today, however, a number of devices are vying for the
title.As ATM grows more popular in carrier networks, so has the idea of using it to
tame wild IP flows. Switched IP, in which ATM devices not only carry IP but
sometimes "think" IP in a whole new way, may turn out to be a critical technology as
voice and data networks begin to move together. This approach, backers say, can
bring discipline to this unpredictable flurry of data.
Switched IP technologies not only speed the progress of IP packets through the
network but, perhaps more important, control the way those packets move from
user to recipient. If carriers can control IP flows--enough to make them
predictable--they can sell a host of services that make little sense on a traditional
data network. Using dedicated IP flows, carriers can deliver voice over IP, fax and
video smoothly. Virtual private networks become possible without elaborate
encryption.
In short, IP networks become much more practical.
Until recently, these benefits seemed out of reach to most carriers, especially
traditional telephony providers with a massive investment to protect. Without a
reasonably firm shared standard in place, many have been skittish about adopting
flow-oriented IP technology.
But many observers believe that the pending Internet Engineering Task Force
standard for IP switching devices, known as multiprotocol layer switching (MPLS),
will prove to be the jump-start the industry needs to introduce the next generation of
IP-based services.
Now that the standards are near completion, carriers are likely to take a closer look
at switched IP, argues Tom Nolle, president of CIMI Corp. "Within the last [few
months], the barriers to IP over ATM have disappeared," he says. "Before then,
there were a number of proposals, but none had any clear constituency."
Doomed to failure?
IP switching made quite a splash a few years ago, when Ipsilon first brought the idea
to the public.
The technology was designed to speed up increasingly choked networks, replacing
slower, more processing-intensive routers with switches.
Ipsilon proposed to provide a software upgrade to the ATM switch that would
make it act like a router. Routers on the other side of the Internet and private IP
network would see the ATM switch as a peer, but the switch would do far less
processing and consequently move faster.
"At the time, a router that could move 5 million packets per second didn't exist, and
if you'd tried to build it, it could have cost a half-million dollars to build," says John
Carosella, vice president of marketing for the company.
Ipsilon was not alone in experimenting with alternatives to router-based networks.
PSINet Inc., for example, built a switched backbone in 1993. Instead of setting up
the standard router-to-router topology, PSINet took the unusual step of plugging its
routers into switches. "Back then, it was kind of a radical idea," says Vice President
of Engineering Mark Fedor.
But Ipsilon's proposal went beyond substituting one device for another. IP
switching, the company later suggested, could help direct packets into individual
flows, whose behavior mimicked that of circuits in the public network. The idea
captured network managers' imaginations.
"I saw Ipsilon present the idea at a trade show back then, and people seemed pretty
impressed," recalls Hilary Mine, senior vice president at Probe Research.
Internetworking giant Cisco Systems later offered its own software-based IP
management approach--tag switching--which, by cutting down the number of
processing routers, would also address the problem of network congestion.
Despite the enthusiastic reception, Ipsilon's idea didn't work. Now a Nokia Corp.
division known as Nokia IP, the former Ipsilon has backed off its original proposal,
arguing that the wide ATM acceptance critical to its plans is unlikely to materialize.
"IP switching based on ATM in the core is unlikely to be broadly embraced by
anybody," Carosella says. "Over time, we've come to the conclusion that it's not a
particularly compelling value proposition."
Others suggest that the Ipsilon technology was doomed from the outset, as it was
unlikely to work efficiently in larger networks. "The Ipsilon strategy required that
every individual user-to-user flow in the network be handled separately in a virtual
circuit," Nolle says. "That imposed a terrible burden on carriers. It just didn't scale."
Today, Nokia IP is creating its own multiservices delivery platform, designed to help
Internet service providers launch new services (Figure 1). "As the Internet grows,
those who carry more traffic don't necessarily make more money," Carosella says.
"The trick is not about making IP go fast, it's about adding value to the traffic you've
got."
A host of other vendors, meanwhile, are taking Ipsilon's place. Though the MPLS
standard is not yet final, at least a dozen vendors have announced plans to release
products based on a preliminary version of the standard. Early players include a
mixed bag of established companies and entrepreneurial start-ups, including
Ericsson, Ascend Communications, Avici Systems and NetCore Systems.
The vendors take varied approaches to switched IP, but the core idea is the same:
to make a "connectionless" data technology behave something like the
circuit-switched network that telecom providers know so well.
What makes these systems different from the previous approach is that they may be
integrating ATM and IP functions in a single device. Also, unlike Ipsilon's original
model, these next generation switches do their processing using chips that are
application-specific, making them faster and tougher than the software-based
product Ipsilon envisioned a few years ago.
One example is NetCore's Everest Integrated Switch, which does both wire-speed
IP routing (up to 2.5 Gb/s) and ATM switching (Figure 2). Everest should begin
trials within the next few months, says John Shaw, vice president of marketing for
NetCore. "What we're trying to do is give service providers a stronger technology
for not just providing a general guarantee of service, but a guarantee on a
per-customer or per-connection basis," Shaw says.
Cisco, for its part, is offering competing solutions that, to some degree, sidestep the
problem. The company began shipping its Gigabit Switch Router 12000 last
October. The next generation device, available this fall, will offer a 2.5 Gb/s
interface that can connect with up to 10 OC-48 (2.4 Gb/s) pipes.
As for sped-up switches, they're just not needed, contends Graeme Fraser, general
manager and vice president of engineering with Cisco's optical internetworking
business unit. "We don't need these new intermediate approaches, as we've scaled
up the capacity of routers to handle huge amounts of data," Fraser says. "The idea
of routers introducing delay is really an old thought."
Who's using it?
While networking and telephony players may have found some common ground, the
IP switching concept is still at a fragile stage. Though vendors are steaming ahead
with IP-over-ATM solutions, they're getting a mixed reception from carriers.
Virtually all the traditional telephony carriers are still hanging back, concerned about
making the costly investment in equipment and just plain conservative about new
networking choices. Most say they're waiting for the MPLS standard to be carved
in stone before they kick the IP switching equipment into production.
Executives with BellSouth, for example, say that despite their interest in offering
tiered classes of service over IP, they're "very standards-based" and don't want to
move ahead until MPLS is a done deal.
BellSouth is starting to offer IP using ATM as a transport mechanism, alongside existing IP over frame relay service. Not far down the road, BellSouth hopes to offer several applications that could depend on defined quality of service (QOS) such as virtual private networking, database access and multimedia, says Barbara
Roden, senior director of network architecture.
BellSouth marketers are eager to get started with the QOS improvements switched IP could bring, but the networking team isn't taking chances. "Our expectation is that
customers will be willing to pay more for better quality of service," Roden says.
"We're certainly hoping to have [switched IP technology] in our labs within 12 months."
AT&T is experimenting heavily with voice over IP and other packet-based applications. For example, the company is planning an ATM-based project mixing voice with data over business access lines, scheduled for launch this fall, as well as a consumer voice-over-IP trial.
When it comes to switched IP devices, the telco is actively looking at Ascend/Cascade's IP Navigator switch, which uses routing software.
But for the moment, MPLS-style switched IP is not part of the rollout, says Bill Leighton, vice president of data network technology. "There's some hardening that has to occur in this technology space," he says. "All of the solutions you see today
are still first generation."
Not every carrier even sees the benefits of an MPLS approach. Some network players are satisfied with the benefits of their existing switched backbone, which can
impose some control on classes of service even if it doesn't create the virtual circuits
envisioned by MPLS.
PSINet, for its part, has kept its existing switched architecture, merely speeding up the connections it serves from T-3 (45 Mb/s) toward OC-3 (155.5 Mb/s) and
OC-12 (622 Mb/s) rates. With that kind of architecture in place, Fedor says, he has little need for an IP switch. "The beauty of what we have is that we can still use the latest routers," Fedor says. "We can control the network without IP switching."
Competitive advantages
Still, newer network service providers, notably ISPs and hybrid network service providers building new networks, say they're likely to try switched IP.
Qwest Communications, for example, expects to build specific overlay networks for specific applications over its IP-only backbone. Without the legacy telephony
network to protect, it's probably smart to work with switched IP, even if the standards aren't yet finalized, says Guy Cook, who served as CEO of ISP Supernet
until it was acquired by Qwest.
"We want to see MPLS come on as fast as it possibly can," Cook says. "But if you're grabbing early market share, there's a lot of importance to just getting out there and getting it done, as opposed to telcos who have a very high revenue base to protect."
Another IP switching fan is Concentric Network Corp., an IP-based network service provider. It still runs the original Ipsilon IP switching equipment in its Santa
Clara, Calif., data center.
Concentric picked up the equipment when it acquired area ISP Internex and has
chosen to stick to that approach.
Internex cut a deal with Ipsilon a year and a half ago and has used that time to bone
up on the pluses and minuses of a flow-based environment. "Yes, we had to spend
on [proprietary] Ipsilon equipment, but the investment turned out to be very
valuable," says Martin Levy, a former Internex exec now serving as Concentric's
vice president of network technology. "We know which flow statistics are important,
how to deal with configurations, what the impact is on ATM switches. That's
something we wouldn't have if we read the glossies today."
Then there are carriers just kicking off major parts of their network architectures.
They could choose to either wait out the standards or jump in.
CLEC Intermedia Communications is deploying its network of ATM switches nationwide, and expects to have about 35 installed by the end of the year. Like its peers, Intermedia is experimenting with switched IP products from several vendors--in its case for use at the edge of its network.
"We're not using it at the core right now," says Bob Rouse, executive vice president of engineering systems and operations. "We don't think there's a need for it. There
will come a time when we need to scale up our IP switching platforms, but it will
happen in 18 months to two years."
However carriers do it, observers say, they'll eventually need to find a way to sort
out high- from low-priority traffic. Otherwise, carriers will have trouble getting the
fees they need, says Jorge Garcia, manager with Arthur Andersen's Advanced
Technology Group.
While it might be more convenient to build a standard Level 2 switched network, the
switched IP hybrid is probably more practical, he says.
"The reality of the marketplace is that IP is wildly popular and that all services could
be based on IP," Garcia says. "In some ways it may be less efficient, but we all have
to pay the IP tax."
Anne Bilodeau Zieger is a freelance writer based in Reston, Va.
Any Comments?
Send them to Karen Murphy at msblues@earthlink.net.
www.internettelephony.com
Telephony July 20
c1998 Primedia Intertec
All Rights Reserved.
| HOME | JOB ZONE | SUBSCRIPTIONS | ARCHIVES | |
