A Horse of a Different Color Integrating ATM and IP
True multiservice capability has long been
the Holy Grail of public networking. A
unified infrastructure with efficient support
for voice, video and all forms of data would
benefit everyone involved.
"At the end of the day, a multiservice
network results in decreased soft-dollar
costs," explains Mike Vent, executive vice president of network
engineering and operations at IXC Communica-tions Inc. "You need
fewer bodies to watch screens, provision services, etc. You also save
money because you need fewer ports. And customers get the benefit of a
single, integrated local loop."
So far, however, the dream has gone unrealized. Integrated services
digital network (ISDN) was simply too little, too late. Asynchronous
transfer mode (ATM) offers superior broadband transport, but it doesn't
mesh easily with existing data protocols. And Internet protocol (IP), the
basis of the Internet, has yet to demonstrate industrial-strength reliability.
Thus, most full-service providers find themselves with parallel switching
fabrics: ATM for high-quality point-to-point services and IP for Internet
access and low-cost connectivity.
Now, the startup community has spawned a collection of equipment
provider aspirants offering gigabit to terabit platforms for wide area
network (WAN) services, some focused on IP-only, but some promising
a new breed of hybrid switches that eliminate the need for dual networks.
By switching both packets and cells simultaneously, products such as the
GigaPacket Node from Argon Networks Inc. or the Everest from
Netcore Systems Inc., combine the deterministic performance of ATM
with the broad application compatibility of IP. Support for multiprotocol
label switching (MPLS) brings circuit-like manageability to connectionless
internetworking, while built-in SONET multiplexing replaces separate
cross-connect devices. Using hybrid IP:ATM switches, network service
providers can--for the first time ever--build practical, profitable
multiservice WANs.
The Network Changes its Stripes
Multiservice capability is essential in today's competitive public
networking market. "You need the multiservices' capability so as not to
force a change in the customer's behavior or systems," says Matthew
Bross, chief technical officer for the Williams Network. "You must bridge
between the customers' legacy systems and the emerging technologies.
Customers' legacy systems represent significant ongoing cash flow, and
you can't just expect them to write these networks off." That is, if you
can't handle all of their traffic, customers may go somewhere else.
And carrying all that traffic with a single, unified network always has
seemed like a good idea. First and foremost, it reduces the number of
devices that service providers must buy, install and manage. The current
confusion of voice switches, frame relay switches, ATM switches,
routers, add/drop multiplexers, digital cross-connects and so forth is
replaced by a simpler, more homogeneous fabric. Capital equipment
costs decrease, and management and training expenses plummet. Service
and revenue growth become far less dependent on the availability of
skilled resources.
At the network edges, customers also benefit from access unification.
With multiple public backbones, each service type requires its own access
circuits: one or more for telephony, frame relay, private lines, etc. But with
only one fabric to access, a single circuit can carry all traffic types. Access
costs--a major component of network expense--drop dramatically. And
new services can be added to the shared links in minutes rather than
months.
Narrowband ISDN was the first attempt at a multiservice architecture.
But ISDN is based on the telephone hierarchy and doles out bandwidth in
64 kilobits per second (kbps) circuits. With local area networks (LANs)
running at 10 megabits per second (mbps) or more, this is way too slow
for most data traffic. And ISDN's end-to-end connections--another
legacy of the phone system--conflict with the hop-by-hop routing model
used in internetworking. Consequently, ISDN found a niche as a
low-speed access solution, but never attained its multiservice goal.
Switching to Packets
Realizing that data soon would overwhelm voice, the telecommunications
industry turned to packet switching. ATM, conceived as a transport
mechanism for broadband ISDN, uses simple fixed-length packets called
cells. The cells travel over virtual connections set up prior to each
conversation. The combination of simple cell formats and virtual
connections (VCs) makes ATM both fast and deterministic.
"Traditional carriers like ATM because it's connection-oriented," says
Bob Bellman, president of Brook Trail Research. "VCs are familiar tools
for managing bandwidth and supporting service-level agreements. Data
folks like ATM because it handles bursty data traffic more efficiently than
TDM (time division multiplexing)."
And everyone likes ATM's high-quality voice and video support. Early
enthusiasts promoted ATM as a universal local and wide area solution.
But ATM came to the party too late. A model of connectionless
communication based on Ethernet LANs and router-based
internetworking was already well established. Connection-oriented ATM
was too different, and the cost and complexity of converting to ATM
were too much for most enterprise users.
Still, for several years ATM was the fastest packet-switching technology
around--the only one that could keep up with synchronous optical
networking (SONET). So service providers adopted ATM as a
broadband transport layer underneath frame relay, transparent LAN
interconnect and a number of other wide area services.
IP Takes Off
While ATM was gaining a foothold in public networks, the World Wide
Web lit a fire under the Internet. Enterprise users turned to IP.
"Since IP was already popular in private wide area backbones," Bellman
explains, "users could exploit public IP services with only minor changes
to existing equipment. Low-cost, distance-insensitive IP-based transport
has become an attractive alternative to leased lines and frame relay." And
Internet service providers (ISPs), with both the technology and the
know-how, have become serious contenders for enterprise traffic.
Now IP is the latest candidate for multiservice immortality. The Internet
provides global connectivity second only to the phone system. IP-based
virtual private networks (VPNs) offer an alternative to both leased lines
and dial-up circuits. Voice over IP (VoIP) promises a low-cost substitute
for conventional telephony. The vision is appealing. And with the advent
of gigabit-speed routing switches, the implementation seems almost
straightforward.
But IP isn't quite there yet. It still lacks battle-proven mechanisms for
prioritizing data flows and controlling delay when the network is busy.
Without these, Fortune 500 CIOs are reluctant to outsource their private
WANs to public IP services, and multimedia support is emerging
technology at best. Moreover, many carriers are uncomfortable with IP's
connectionless architecture for high-revenue VPN services. How can they
provision bandwidth accurately or ensure compliance with service level
agreements (SLAs) if they can neither monitor nor manage individual
customer flows?
Torn Between Two Layers
Given that neither technology offers a perfect multiservice solution--at
least not yet--most service providers choose to operate dual networks.
The IP network supports applications such as Internet access and dial
VPNs, for which cost or connectivity are important. The ATM network
supports frame relay VPNs, circuit emulation, private branch exchange
(PBX) interconnect and other applications where reliability and quality of
service (QoS) take priority.
"Both IP and ATM are absolutely essential," says the Williams Network's
Bross.
"ATM delivers the predictability and manageability that enterprises
demand, while IP fosters all sorts of innovation at the edges," agrees
IXC's Vent. "IXC is agnostic. Data is good--ATM or IP or both.
Admittedly, IP is going to be a huge chunk of the communications
business, but you can't force customers to go a certain way."
The dual approach usually entails multiple overlays. The bottom layer
delivers raw SONET/synchronous digital hierarchy (SDH) bandwidth,
perhaps over a substrate of dense wave division multiplexing (DWDM).
ATM rides on top of SONET, supporting frame relay, circuit emulation,
voice traffic and native cell relay. The IP layer interconnects routers using
direct SONET links, ATM virtual circuits or a combination of the two.
The multiple overlay model is easiest and least expensive in cases which
an ATM infrastructure is already in place or where an ISP is buying
bandwidth from a wholesale provider.
Several mechanisms have, however, been invented for linking IP more
closely to ATM, including Ipsilon's IP switching, Ascend's IP Navigator,
the ATM Forum's multiprotocol over ATM (MPOA) and the Internet
Engineering Task Force's (IETF's) MPLS. By identifying individual IP
flows and tying them to ATM VCs, all of these "IP switching" schemes
exploit ATM's speed, QoS and manageability. Of all the methods
proposed however, only MPLS--soon to be standardized by the IETF--
is designed specifically for WAN and VPNs.
MPLS employs standard routing protocols such as open shortest path
first (OSPF) to lay out Layer 2 switched paths through the network. As
packets enter the network they are assigned labels that represent the
paths. Nodes within the network forward each packet based on its label
rather than its destination IP address. Labels can be ATM VC identifiers,
frame relay data link connection identifiers (DLCIs), or special IP "shim"
headers for packet on SONET (PoS), making it easy to integrate ATM,
frame relay and IP in a multiservice backbone.
By relating each flow to a specific path through the network, MPLS
brings connection-like manageability to connectionless IP. An
MPLS-labeled path acts as a tunnel to keep VPNs separate from one
another. Enterprises with extensive use of the "private" net10 IP address
space can use a public backbone without renumbering all of their
endstation addresses or using difficult network address translation (NAT)
devices. Only the routers at the public/private interface--the same routers
that add labels to incoming packets--need public IP addresses.
Reducing Grooming
In most WANs today, raw bandwidth is provided by a system of
SONET rings and add/drop multiplexers. At each customer connection
site, multiple bit stream subchannels are "groomed" onto the SONET ring
and build an aggregate SONET channelized link transporting multiple
customers' traffic across the WAN. At central offices (COs) with service
platforms such as ATM switches or IP routers, a digital access
cross-connect system (DCS) breaks out the sub channels or "degrooms"
them over separate physical connections to the appropriate service
platforms.
Hybrid IP:ATM switches with integrated SONET multiplexing support
offer a cost-effective alternative to the multiple overlays and associated
grooming/degrooming DCSs. Each hybrid node unites ATM switching
with IP routing in an integrated package. Any interface in its entirety or at
the SONET subchannel level can be configured to support ATM, IP or a
combination of the two. Thus, an interface to the hybrid node may carry a
single high-speed flow of either IP or ATM traffic, or it can carry several
separate flows using standard SONET framing. Some channels may carry
IP POS, some may carry ATM cells, some may carry IP over ATM and
so forth.
"If you want to remain agnostic around IP and ATM," Bross says, "you
want to pick the optimal type of trunking for each customer, either IP or
ATM, from the OC-x bundle."
With IP:ATM support and integrated SONET multiplexing not only are
there fewer interface cards to buy and manage, but DCSs can be
eliminated, and the individual management system for each previously
separate platform--the ATM switches, frame relay switches, IP
routers--is also eliminated. With hybrid IP:ATM switches, service
providers save money because they have less equipment to buy, install
and manage. And because the service configuration is modified via
keystroke, not hardware swapping operations, hybrid switches adapt
quickly, easily and cost-effectively as the mix of IP and ATM shifts. |
