Avalanche of Terabit Routers Means Confusion for Engineers

Ever more vendors are moving to address the explosion in
data traffic through a diverse array of terabit-router strategies that are sure to make
life harder for engineers before it gets easier.

Cisco Systems Inc. is the latest major vendor to commit to
the terabit-router class, adding still another type of solution for network operators to
sort through in this still-evolving and largely nonstandardized product arena.

After holding off on plans for such gear as other vendors
promoted their strategies through the second half of last year, Cisco chose the week of
the Western Show to quietly let the world know it would join the stampede.

"We expect the cable industry, along with the folks
who are deploying DSL [digital subscriber line] networks, to be the driving force in
demand for terabit routers," said Paul Bosco, vice president and general manager of
Cisco's cable and wireless business unit.

"Take any large metro area today and you find the
networks operating with huge OC-48 rings dedicated to data traffic," he added.
"As you add video and voice to those rings, the volume adds up really fast."

Cisco's strategy is based on operator deployments of
what Bosco called the "GSR Plus" version of its gigabit-switch router -- a
320-gigabit-per-second capacity system formally known as the "GSR12016," which
the vendor has just begun to ship.

Cisco's "12000 Terabit System," to be made
available later in the year, uses an advanced crossbar-switch fabric to tie multiple GSRs
together at a single point of presence, creating a single routing point that can scale up
to 5 terabits per second.

The result is a fully integrated terabit router, where
capacity is shared seamlessly across all of the line cards without causing interruptions
when the traffic requirement exceeds the capacity of any one element, Bosco said.

The GSR12016 marks Cisco's first use of an OC-192 (10
gbps) interface card in its routers, and it is a key step on the way to eventual
introduction of all-photonic routing by wavelengths.

In this scenario, using technology from the
"Monterey" product line Cisco acquired with its purchase of Monterey Networks
Inc., total throughput would eventually scale to 320 tbps.

While Cisco has been a dominant force in Internet-protocol
routing at the GSR level, competitors believe they can persuade network operators that the
move to terabit routing requires more than merely scaling up from the Cisco architecture

Lucent Technologies will be bringing out terabit routers
through its recently acquired Nexabit Networks unit in March, which take a very different
approach to scaling beyond the GSR level.

"Everybody is looking for massive scalability, but the
gap between the performance of optics and electronic switching is a problem," Nexabit
CEO Mukesh Chatter said. "What we bring to the table is a 6.4-tbps switching capacity
per chassis, which is the highest switching capacity per chassis anybody has ever

Combined with low latency of 40 microseconds "through
the box" and high port density that can accommodate individual ports running at
interfaces of as low as OC-3 (155 megabits per second), the high switching speed of the
new Lucent router provides the quality-of-service support essential for scaling standard
IP traffic and for making voice over IP a reliable component of that traffic. "This
is the enabler for voice or real-time traffic in the core of the network," Chatter

But terabit routing brings into play the need to provide
for network restoration in the optical layer of DWDM (dense-wavelength-division
multiplexing) systems at the point of interface with the router. Here, the
incompatibilities among vendor solutions are rampant, making it virtually impossible to
use more than one vendor's terabit-router system in any given network.

"As you start integrating DWDM with terabit-class
routers, [network restoration] becomes a major issue," Chatter said.
"Intellectual property has to be respected and standards have to evolve together, and
that's still a pretty fuzzy area."

Lexington, Mass.-based start-up IronBridge Networks is
applying an internal optical-interconnection system that will come into play as a factor
in wavelength routing later on, according to vice president of marketing Doug Antaya.

"We're not talking about our optical architecture
in any great detail, but you can assume it has implications for wavelength routing,"
he added.

The internal optical system is meant to provide a seamless
means of scaling traffic from low speeds to terabit speeds, Antaya said, adding, "To
get to terabit speeds, you need to link shelves together in a way that is transparent to
the outer network, and that requires an optical interconnection internally."

Linking internal and external optical technology requires
using proprietary means of wavelength restoration at the core of the router. While this
renders SONET (synchronous optical network) redundant in many respects, it also locks the
supplier into a system that may or may not be compatible with whatever standards emerge
for optical-layer grooming in the future.

"With SONET, there's a fair amount of hardware
required for the framing chips, which is something we'll supply to accommodate the
tremendous amount of installed networking systems that are out there today," Antaya
said. "But we'll also supply other versions of network cards that avoid SONET,
so that you can, for instance, operate in gigabit Ethernet end-to-end as a low-cost
alternative to SONET."

Where an OC-192 card for a router might cost $200,000, a
10-gbps Ethernet card will cost $20,000 to $40,000, Antaya noted. Support for network
restoration will require a uniform approach across all routing points, making it necessary
to rely on one vendor's system, he acknowledged.

Optical-layer incompatibilities are but one of the many
areas of distinction among the providers of terabit routers. Most other differences apply
to how the systems work internally, without any effect on the ability to interoperate with
other vendor systems.

These differences have to do with how much signal
processing is done in hardware versus software, the functions performed by
packet-forwarding engines, how much buffering is employed in the system and many other
elements of system design.

Here, the challenge for network engineers is to figure out
whose claims are valid and which processes are best-suited to future scaling needs at a
time when one can only guess just what those needs will be.

Lucent's approach to helping operators sort through
these questions is to avoid taking a rigid stand on the specifics, noted vice president
for cable communications Dee Dee Nye. "We want to give you options that are
specifically geared to your needs without getting involved in the religious wars,"
she said.

The role of terabit routing will be defined over time, with
plenty of support in the meantime for non-IP approaches to traffic management, she noted.

Bosco made much the same point, noting that along with
offering a flexible, scalable approach to routing via the GSR-to-Terabit 12000 migration,
Cisco is providing the tools operators require for operating in the ATM (asynchronous
transfer mode) and TDM (time-division multiplexing) domains.