Seeing the Light
Optical switches will be the next big thing in data transmission
By Bill Alpert
The optical Internet is a modern wonder. In the last decade, the cost for a
kilometer's worth of gear that carries a billion bits per second of data has
plummeted -- from $1,000 to $100. Firms like Nortel Networks and Ciena
delivered that deep discount, with gear that packs dozens of signals on a single
optical fiber, overlapping different colors of light through the technique called
wavelength division multiplexing. By doubling their price-performance every 10
months, optical networks have left semiconductors in the dust. Moore's Law, as
every kid knows, doubles computer processor power just every 18 months.
"Bandwidth is cheaper than processing power," said Nortel chief executive John
Roth on Thursday, at the 25th anniversary of the Toronto firm's New York
Stock Exchange listing. The Big Board's trades, Roth proudly noted, are
crossed on computers in Brooklyn, not Wall Street.
Unfortunately, all that bandwidth is not good enough. Internet traffic is doubling
every 100 days. While fiberoptic transmission of that traffic has gotten cheaper,
the electronic switches that direct the traffic are doubling their
price-performance only once every two to three years. That's why the next
technological leap in optical networks has to be optical switches. The same day
as Nortel's gathering, Vancouver-based 360networks announced plans to buy
as many as 100 optical switches from Sycamore Networks. Chelmsford,
Massachusetts-based Sycamore is just one of a dozen vendors working to
perfect optical switches. Before this fall's collapse in tech stocks, optical
switches were a primary cause of Wall Street's mania for optical stocks like
Sycamore and Corvis.
A stock like Corvis never merited its August price of nearly $115, which valued
the Columbia, Maryland, firm at $38 billion-some 125 times its hoped-for
revenues next year. But market researchers predict that optical switching will
be a $5 billion annual business by 2004, and even in the now deflated
environment for tech stocks, a number of optical switch firms have lined up for
their initial offerings.
All the burgeoning content of the Internet -- from e-mails to videos -- is made
of electrons, notes Cisco Systems spokesman Kent Jenkins.
Cisco, of course, made the big time by
figuring out how to route all those
electrons. The routing switches of
Cisco, and such challengers as Avici
Systems and Juniper Networks, now
shuttle those electrons around at speeds
of 2.5 billion bits per second.
Nortel, however, has been outfitting the
backbone of the Internet with lightwave
multiplexing gear that runs at 10 billion
bits per second. Next year, Nortel is
cranking the speed up to 40 billion.
This is possible because of the
wavelength division multiplexing pioneered by Linthicum, Maryland-based
Ciena. The WDM technique takes electrons off copper wire and converts them
into lightwaves beaming across fiberoptics. The photon particles of light move
faster than electrons, of course. Better yet, photons can pass right through each
other, allowing WDM to simultaneously send separate streams of light down the
fiber, each stream of a slightly different wavelength (or "color"). Nortel gear
packs 160 wavelengths on a fiber, with a 320-channel system in the works.
When those streams of light need to make a turn on the superhighway, or get
off at their final destination, they've got to be converted into electrons-because
until recently, switching gear has been electronic. To keep up with just one
channel of photons barreling at 10 billion bits per second, therefore, it takes four
lines of electronic switching. As that disparity gets wider, it gets more
cumbersome and expensive for electronics to keep up.
"It's like building an interstate highway with no developed roads off the exit
ramps," says Karen Liu, an analyst at market researcher RHK Inc. "You have
huge traffic jams."
Keeping up isn't the only motivation for optical switching. Global Crossing,
Qwest Communications and Williams Communications, all of which have
newfangled networks, figure that optical switching will greatly reduce the cost
of building and operating the Internet.
The world's communications networks were built for voice traffic, which runs in
steady, small pieces. A wonderfully fault-tolerant technology called SoNET
carried this traffic along doubled-up rings of fiberoptics. If one ring got cut,
electronic connections shifted traffic to the spare ring in an imperceptible 50
thousandths of a second.
Data traffic has now overtaken voice traffic on many networks, and the
evolutionary legacy of networking gear has resulted in roadblocks.
Data traffic undergoes several levels of repackaging in order to get on and off
the 'Net's optical backbone. Packets get addressed in a protocol called IP, then
bundled for express delivery in yet another protocol dubbed ATM. Next, they're
converted again for transmission over failsafe SoNET rings, then finally
packaged by WDM gear as wavelengths for the network's backbone.
The great hope for directing this traffic once rested on ATM switches.
However, converting optical traffic into ATM's electronic format is like taking
passengers off a high-speed express and herding them single-file onto another
train headed in the direction they want to go.
In contrast, optical switches, in effect, just turn the whole express train around
and send it and the passengers on the correct route. They do this by switching a
stream of optical data without going through conversion to other protocols.
By collapsing today's four-level hierarchy of data formats into just two, optical
switch vendors promise to generate big savings for their customers.
Tellium, an Oceanport, New Jersey, firm now in registration for its initial public
offering, can replace 40 racks of SoNET-style gear with four racks of optical
switches, asserts Chief Executive Harry Carr.
New carriers like 360networks and Level 3 Communications are using optical
switches to avoid the inefficiency of SoNET's paired-ring structure, which
leaves half of a network's capacity idling to ensure sub-second failovers. (In
other words, if one part of the system crashes, the other will almost
instantaneously take over, with no data loss.) A mesh of optical switches, in
comparison, might deliver fault-tolerance with less than a quarter of the fallback
capacity.
Ironically, most "optical switches" today are electronic at their core. Firms like
Ciena, Sycamore and Tellium can take in wavelengths streaming down the
channels of a WDM fiber and switch them out to another WDM fiber-the
traffic's converted to electrical form for switching in the guts of their devices.
"The industry's done a disservice to the English language by using the term
'optical switch' for something that's electronic in the middle," chuckles Don
Smith, president of Nortel's optical Internet unit, which itself sells an
optical-electrical-optical product called the OPTera Connect DX.
In contrast to such "OEO" devices are all-optical switches that redirect
lightwaves in original form. Smith and others call such a product a "photonic
switch" or just "OO." A debate has raged between proponents of OEO and
those of OO.
The OO device, in fact, is a key part of the strategy of Corvis, which came
public in July at the peak of Wall Street's optical frenzy, raising over $1.1 billion
through the sale of 32 million shares at $36 apiece. With more than $550 million
in contracts in hand from Broadwing Communications, Qwest Communications
and Williams Communications Group, Corvis saw its shares crest near $115 and
then dive with the rest of the network sector, falling below $20 before
recovering a bit to a recent $32.
With a portfolio of all-optical technologies, Corvis boasts that it can save its
customers 90% of the cost of building a network. The firm has intrigued -- and
maddened -- outsiders by refusing to disclose details of its technologies, even to
the brokerage firm analysts that tout Corvis shares. Fans find reassurance in
the credibility of Corvis' founder, David Huber, who was the brains behind
wavelength multiplexing at Ciena.
"It's a technological black box," maintains Shyam Jha, Corvis's vice president
for marketing. "But it's a black box that works." Corvis justifies its
secretiveness by claiming it has an 18-month lead over Nortel and other
optical-network rivals and doesn't want to reveal anything that might help them
catch up.
Jha will say only that the Corvis device can switch a stream of 960
wavelengths, traveling at 2.5 billion bits per second.
Among the few who have seen the switch are Corvis customers.
Engineers at both Williams and Broadwing have told Barron's they've been
happy with what they've observed. Chris Rothlis, the VP who runs Broadwing's
optical lab in Austin, Texas, says his company has two switches. It's now
installing them in its network for field testing.
Jha says that his firm will reveal its technology as customers turn on the
switches. Investors will no doubt take heart, although with a $10 billion stock
market value, Jha's company still is far from cheap. Expected revenues are $30
million this year, and $300 million in 2001, with breakeven expected to arrive in
2002.
Other all-optical switch contenders have been more open about their
technologies. Indeed, a bewildering variety of switching fabrics have been
proposed, ranging from swiveling arrays of tiny mirrors to ink-jet-style bubbles.
The latter approach is being explored by the Hewlett-Packard spinoff Agilent
Technologies, with the encouragement of French telecom supplier Alcatel.
Lynx Photonic, a privately held Israeli firm with operations in Woodland Hills,
California, is pursuing switches built of a material called lithium niobate, which
appears to be very fast and very immature. Another Israeli startup, Trellis
Photonics, aims to bounce lightwaves off holograms embedded in crystal
arrays.
Chorum Technologies, a Richardson, Texas, firm in registration to raise as
much as $150 million in its IPO, has delivered prototype components that use
liquid crystals like the display of a laptop computer. Backed with $158 million in
venture funding, Chorum had $3.3 million in initial revenues in the September
quarter, and a loss of $44 million. Other component suppliers for solid-state
photonic switching include PMC-Sierra and Applied Micro Circuits. The most
hotly pursued technology for all-optical switches, however, are
microelectro-mechanical switches, or MEMS. These are arrays of movable
mirrors cut from a wafer of silicon.
Lucent Technologies made a splash with the announcement of the first
MEMS-based switch, the LambdaRouter. Global Crossing is using it to switch
traffic between New York and England.
In March, Lucent rival Nortel paid over $3.2 billion in stock for MEMS-maker
Xros. Greg Reznick, who runs Xros, plans to deliver a switch next year that can
switch 1,152 links. The beauty of all-optical switching, says Reznick, is that it
can switch signals in any format, wavelength or speed. In contrast, OEO
switches have to upgrade their electronics every time the network speeds up.
To scale up to 1,000 links, the mirrors in switches like those of Lucent and
Nortel must be able to swivel along two axes, like a ship's compass. Tellium, the
switchmaker in registration for a $260 million IPO, bought comparable MEMS
technologies this year, to supplement the 512-link OEO switch it's already
selling.
Also in registration for an IPO is San Diego-based OMM, which sold $1.5
million in MEMS components to customers like Alcatel. OMM's mirrors only
flip along one axis, however, and it must develop "3D" technology to supply
large switches.
Ultimately, says Tellium's Harry Carr, optical switches will need both mirrors
and electronics. When fiber channels commonly run at 40 billion bits per
second, in a couple of years, only all-optical will be able to keep up. But outside
the core of the network, those lightwaves will have to be decoded into electrons
for delivery to a network's customers.
E-mail comments to editors@barrons.com |
