Happy Together
IP and ATM proponents have suddenly hopped on the peace train
SUSAN BIAGI
ut on your love beads and join the peace brigade. The once-fervid battle between
Internet protocol and asynchronous transfer mode appears to have reached a cease-fire.
Parties from both camps are holding up white flags, opting for peace among the
protocols.
The reality is that both IP and ATM are here, and that isn't changing anytime soon.
Vendors and carriers agree that they need to find a way to make the two protocols work
together.
The overall attitude--from companies on both sides--is that each fits in a specific place
in the network. IP is used for access; ATM fits in the core. Although some claim that
eventually any technology will work in the core, especially as we become further
entrenched in optical networking, ATM is the preferred choice today because of its
traffic management and quality of service (QOS) features.
Positive vibes for ATM
"It's clear that IP is becoming a very important technology. But all the major [Internet
service providers] are using ATM as their core switching technology," says Steve
Vogelsang, director of technology at Fore Systems.
One reason is that ATM switches have very high capacity. For example, Fore offers a 10
Gb/s switch today, and a 40 Gb/s version will be available in August. Nortel Networks
markets the Passport 15000 multiservice switch, which scales from 40 to 160 Gb/s. And
start-up ATM switch vendors such as Argon Networks, Avici Systems and Juniper
Networks are creating similar high-capacity products under the terabit switch/router
heading.
IP routers are just beginning to reach those marks, Vogelsang says. For instance,
Ascend Communications' GRF products have only a 4 or 16 Gb/s switch fabric. But
Cisco Systems sits at the top of the router game, and the company's new family of
gigabit switch routers is designed for service providers' growing needs. Cisco's 12000
GSR ranges from 5 to 60 Gb/s.
Service with a smiley
Service aspects carry at least as much weight as capacity. Carriers--and customers--are
attracted to ATM's multiservice capabilities and advanced traffic engineering
capabilities.
"ATM is a trusted and true technology," says Deb Pierson, executive manager for ATM
product management at MCI WorldCom. "We're seeing 100% growth year over year on
[our] ATM network." Many ISPs use ATM switching because of its QOS guarantees,
she says.
Unlike IP's best-effort service, ATM offers built-in QOS levels so service providers can
define service level agreements and guarantee delivery. That capability generates
additional revenue for carriers and ISPs, which can then provide--and charge for--more
tailored service offerings.
"ATM gives ISPs and service providers the ability to manage and optimize the way
traffic flows across the core network," Vogelsang says. "ATM provides what IP isn't
quite ready to do."
Not quite, but telcos are taking strides to give IP some QOS capabilities. Differentiated
services and multiprotocol label switching (MPLS) are two standardization efforts.
DiffServ replaces a previously ignored field in the IP header with the "type of service"
field that defines a few service levels. It's gaining momentum, but MPLS, with broader
management capabilities, is turning more heads.
MPLS, born of Cisco's tag switching concept, translates IP header information into a
label that ATM switches can read. The label identifies the type of traffic, path and
destination. Unlike traditional routing, in which packets are forwarded on a per-hop
basis, MPLS enables more direct routing based on factors such as traffic type or
destination.
In the happy land of protocol-agnostic networks, MPLS will play a huge role, says Rob
Redford, director of product marking in Cisco's multiservice switching business unit.
MPLS will support IP or ATM at the core, and a host of services such as virtual private
networking. Because Cisco believes that future networks will be built on MPLS, the
company created its IP+ATM product line.
In the IP+ATM model, the IP capability sits at the edge and IP or ATM fits in the core
to support additional service such as ATM leased lines. It's a more simplified approach
than running IP over ATM and allows for multiple services to run on one platform,
Redford notes (Figure 1). In the case of IP, Cisco's MPLS implementation eliminates the
need to translate IP routing and addressing information into a label that ATM can read,
he says, adding, "ATM switches natively use the same label-swapping paradigm
employed by MPLS, making MPLS a natural fit for ATM switches."
MPLS is the key to giving ATM visibility into IP packets. "For service providers, ATM
makes a lot of sense because they need to offer multiple services: frame relay and ATM
for today, and IP as they build out," Redford says.
Fore's solution is "capacity-aware routing" which adds a traffic engineering function to
private network-to-network interface (PNNI), Vogelsang says. Fore claims that a proper
IP-over-ATM implementation uses a layered architecture (Table 1). An extendible
service layer, the IP routing part, must be able to address billing, provisioning and
accounting issues. Linked via MPLS, the switching layer is part of the ATM
infrastructure. It handles the transport and traffic engineering capabilities. All of that
rides on the transmission layer.
Fore's value-add comes in path selection process, says Vogelsang. Using a weighted,
unspecified bit-rate, service providers can load-balance their networks by changing the
traffic path. If the most direct path has 80% usage, a less populated indirect route might
be faster (Figure 2).
DiffServ and MPLS are ways to bridge the gap between ATM and IP, but their
deployment is limited because standards have not been finalized. With an eye toward
the future, carriers continue to maintain separate networks, each optimized for a different
purpose.
The carrier lovefest
Carriers have remained neutral in the IP/ATM lovefest, offering both technologies.
Although it sounds admirably altruistic, it's simply driven by the need to serve customer
demands.
"We at Qwest are not trying to [force] ATM or IP," says Ian Dix, a vice president of
data marketing at Qwest Communications. "We're focused on the application, what the
customer is trying to achieve." Qwest maintains separate frame relay, ATM and IP
networks, but it has combined the network operations center, element management and
network management capabilities for its IP and ATM networks in an effort to lower the
cost of provisioning and grooming.
The choice to use IP or ATM as a transport mechanism hinges on QOS and cost
tradeoffs. Running IP over ATM results in the cell tax, which poses potential cost
inefficiencies, says Guy Cook, a vice president of data marketing at Qwest. Still, the
carrier guarantees its IP service even though IP doesn't contain tracking mechanisms.
How? Qwest's answer is to open more pipes.
An excessive amount of bandwidth will stave off throughput problems, but it's an
expensive way to maintain a network. And critics note that boosting bandwidth on an IP
network doesn't address the issue of guaranteed delivery. "If you're just throwing
bandwidth at the problem, you not fixing the differentiated services problem," says
Chris Nicoll, senior analyst with Current Analysis.
Qwest counters that its networks are not dropping packets. "We're not going to run into
bandwidth issues," says Dix. Both the IP and ATM networks "have what we need to
handle the traffic, and we still only use 50% of our capacity."
Like Qwest, MCI WorldCom sees the need for both IP and ATM but has taken a
different approach. Last fall, MCI WorldCom announced its On Net concept that
converges IP, ATM and frame relay.
The customer focus is on applications, Pierson says. Customers don't want to know all
the details about the ATM or IP service, they just want to be able to broadcast the
CEO's speech to five locations. "Customers are looking into converging voice, video
and data into one network. That's not anything new. That's the promise of ATM," she
says.
Bell Atlantic also maintains networks with IP and ATM devices, says Ray Albers, vice
president of network architecture for the carrier. "We like ATM for the next few years
for its ability to transport multiple protocols," he says. Switching manufacturers are
starting to move to multiprotocol switching anyway, he adds.
In terms of convergence, "IP is where the action is for the Internet and moving data
around. But as we sit up nightly worrying about how to migrate from a circuit-switched
orientation to a data orientation, ATM looks more like the viable technology for the next
two to five years."
Building data-optimized networks is a Catch-22, Albers points out. IP might be the wave
of the future, but right now, "part of the problem is that data is 90% of the traffic but
only 10% of the revenue."
Coexistence still rules the day, says Mike Arellano, an analyst with Degas
Communications Group, adding that customers don't yet need one unified ATM/time
division multiplexed (TDM) network. "Big customers have needs for ATM pipes and
TDM pipes. Why just be specialists in IP? Customers don't care that you're the leader in
IP as much as they care that [they] have a certain set of needs and problems, and can
you fix them," he says.
Despite the peacefully mellow attitude of vendors and carriers, the press continues to
fuel the IP vs. ATM debate. "It's a heated debate, but it doesn't warrant the level that's
out there," says Bryan Van Dussen, vice president of marketing with Cignal Global
Communications. Cignal Global maintains the SM-10 network, a "switched multimedia
transmission network" based on ATM. Target customers are ISPs, carriers and large
businesses.
"In three or four or five years, both protocols will find an adequate place in the next
generation carrier," Van Dussen says. In fact, he says, "emerging carriers are using
ATM to develop IP services. That's the same thing we're doing."
Despite the initial hype, few expect to see ATM go to the desktop.
"If it's going to the desktop, it's most likely going to be IP to the desktop and ATM as
the wide-area technology," says MCI's Pierson.
Jerry Norton, vice president for site development at Delta Three, agrees. Delta Three
runs IP over frame relay and maintains IP at the edge for voice-over-IP services. The
company doesn't see a push for ATM at the edge. "So far we don't see it as a fight. We
are IP," he says.
Delta Three enjoys the connectionless orientation of IP, but "more key is the
development of value-added services," Norton says. The talent pool of programmers for
IP-based solutions exceeds that for ATM, giving rise to more innovation, he argues.
Despite the QOS advantages of ATM, "we have the view that it is harder to deal with
equipment, and fewer people know how to run it at the edge," he says. But IP has its
problems, too, he concedes. For example, switched virtual circuits don't scale on an IP
network.
Above all, he says, "the alternate core network is IP-based. We're not jumping out there
on the core and the edge bandwagon. There is no need to declare what we want for the
core, and IP is working well out at the edge."
The world will be beautiful
Service providers expect change but are reluctant to move too quickly. Substantial
investments in legacy ATM equipment coupled with its proven track record make it
hard to let go.
As Nicoll observes, "If you have to carry any circuit-switched traffic at all, the best way
to do that is over ATM."
The economic impact is certainly a question, too. Carriers have to maintain revenue
while building out for the future. Despite the IP hype, little information exists on the
economics of building IP networks, says Ted Sienicki, director of sales and marketing at
Probe Research. At first glance, the process of bringing IP QOS up to an acceptable
level appears to be "complicated and expensive," but no concrete data exists, he says.
Probe is researching the network costs, operational issues and related costs of
improving QOS, integrating voice and data, and adapting circuit-switched networks to
support increased data traffic. The study will analyze the costs of building new IP and
circuit-switched networks, and compare that with the existing public network.
"IP won at the application or convergence layer," declares Nicoll. "It hasn't won at the
transport layer universally, primarily because of QOS. There's not one clear winner out
of all this. I don't think the only network model you'll see is IP. In the future, [we'll see]
IP over optical."
Running IP directly over dense wavelength division multiplexed channels would
eliminate the ATM cell tax, but that's a bit premature. "The IP zealots preach that IP is
ultimately going to IP over glass," says Bell Atlantic's Albers. "That's probably an OK
direction in the long term, but it neglects things like network management, restoration
and reliability. There's a lot of work there."
Fore's Vogelsang echoes that sentiment. "If we could run IP over photons--discard all
the transmission technology and discard all the switching technology, and just put IP
packets onto light--the world [would] be beautiful," he says. "That's a good long-term
view, but the delta between here and there is quite large. It's a gulf."
But if that's the only discordance in the peace-loving environment of IP and ATM,
you're sure to have a smile on your face.
The edge of Aquarius
SUSAN BIAGI
Like most vendors in the new harmonious world of Internet protocol and asynchronous
transfer mode, upstart TransMedia Communications is serving up both in the interest of
customer needs.
The company's tag line is "creating the new edge and sharpening the core," and
TransMedia's approach is to go for the service angle. Regardless of the age, size or type
of network, all service providers want to do just that: provide services. The protocol just
gets in the way.
"ATM is deployed by existing carriers," says Randy Brumfield, director of product
marketing at TransMedia. "Newer telcos want to put in a single IP network to manage
vs. multiple [networks] or put in enough bandwidth in the backbone to offset the
inefficiencies that IP has with [quality of service]."
Building an IP network that will provide QOS is expensive--economically unfeasible for
smaller companies, and hard to justify for larger ones. "If you don't have gobs of
bandwidth out there, it's hard to guarantee [service]," says Brumfield.
So how do you marry IP to the ATM backbone? Brumfield's answer is the MMS-1600, a
technology-agnostic edge switch that terminates circuit-switched traffic for transport
over IP, ATM, frame relay, time division multiplexed networks and the public network
(see figure). It's a 5 to 10 Gb/s "media-aware" switch that can route calls based on
source and destination, service level or network status.
TransMedia claims that by confining their focus to the Internet, emerging carriers are
losing potential revenues from legacy traffic. Multiple networks are expensive to
maintain and hard to manage, and the lack of integration makes it tougher to roll out new
services. The MMS-1600 handles multiple services, acting as a gateway to the
backbone. And it's wide open for developers.
"If you divorce [the switch] from the platform, you open up the door for a whole set of
programmers to develop applications," Brumfield says. Once the service platform is fully
open, programmers will flood the market with new service applications, he says. And the
big winner in this scenario is the service provider.
The open-application policy has validity, but no one knows to what extent.
"Applications for IP will grow faster, but it's hard to measure," says Mike Arellano, an
analyst with the Degas Communications Group.
For questions or comments,
contact Web Editor Karen Murphy msblues@earthlink.net .
Please report any problems with this site to:
Webmaster@internettelephony.com.
www.internettelephony.com
Telephony February 1
©1999 Intertec Publishing Corp., a PRIMEDIA company
All Rights Reserved.
| HOME | JOB ZONE | SUBSCRIPTIONS | AR |
