I don't know if this is of interest to anyone. I just wondered
how ADAP would tie into this.
Was it Britney Spears or Fatboy Slim? The network
administrators at Kent State University had not a
clue. All they did know last February was that
"Rockefeller Skank" and hundreds of other
downloading hits had gotten intermingled with
e-mails from the provost and research data on
genetic engineering of E. coli bacteria. The university
network slowed to a crawl, triggering a decision to
block access to Napster, the music file-sharing
utility.
As demand for network capacity soars, the Napster
craze may mark only the opening of the first of many
floodgates. Venture capitalists, in fact, have
wagered billions of dollars on technologies that may
help telecommunications companies counter the
prospect that a video Napster capable of
downloading anything from Birth of a Nation to
Rocky IV might bring down the entire Internet.
PowerPoint slides at industry conferences emphasize why the deluge is yet to come. Video
Napster is just one hypothesis. A trillion bits a second--the average traffic on the Internet's
backbones, its heaviest links--may fulfill less than a thousandth of future requirements. Online
virtual reality could overwhelm the backbones with up to 10 petabits a second, 10,000 times
more than today's traffic. (A petabit is a quadrillion bits, a one with 15 trailing zeros.) Computers
that share one another's computing power across the network--what is called
metacomputing--might require 200 petabits.
If these scenarios materialize--and, to be sure, people have been tapping their feet for virtual
reality for more than a decade--the only transmission medium that could come close to meeting
the seemingly infinite demand is optical fiber, the light pipes trumpeted in commercial interludes
about the "pin drop" clarity of a connection. Fiber links can channel hundreds of thousands of
times the bandwidth of microwave transmitters or satellites, the nearest competitors for
long-distance communications. As one wag pointed out, the only other technology that comes
close to matching this delivery capacity is a panel truck full of videos.
The race to augment the fiber content of the world's networks has started. Every day installers lay
enough new cable to circle the earth three times. If improvements in fiber optics continue, the
carrying capacity of a single fiber may reach hundreds of trillions of bits a second just a decade or
so from now--and some technoidal utopians foresee the eventual arrival of the vaunted petabit
mark. To break that barrier, however, will require both fundamental breakthroughs and the
deployment of technologies that are still more physics experiments than they are equipment ready
to be slotted into the racks on nationwide phone and data networks.
New photonic technologies, which use lightwaves instead of electrons for signal processing, will
make current electronic switching systems obsolete. Even now the transmission speeds of the
most advanced networks--at 10 billion bits a second--threaten to choke the processing units and
memory of microchips in existing switches. As the network becomes faster than the processor, the
cost of using electronics with optical transmissions skyrockets. The gigabit torrent contained in a
wavelength of light in the fiber must be broken up into slower-flowing data streams that can be
converted to electrons for processing--and then reaggregated into a fast-flowing river of bits. The
equipment for going from photon to electron and back to photon not only slows traffic on the
superhighway but makes equipment costs soar.
While network designers contemplate the prospect of machine overload, hundreds of companies,
big and small, now grapple with creating networks that can exploit fiber's full bandwidth by
transmitting, combining, amplifying and switching wavelengths without ever converting the signal to
electrons. Photonics is at a stage that electronics experienced 30 years ago--with the development
and integration of component parts into larger systems and subsystems. A rising tide of venture
capital has emerged to support these endeavors. In the nine months of 2000, venture funding for
optical networking totaled $3.4 billion, compared with $1.5 billion for all of 1999, although this
may have eased in recent months. The success of a stock like component supplier JDS Uniphase
stems in part from the perception that its edge in integrated photonics could make it the next Intel.
Investment in optical communications already yields payoffs, if fiber optics is matched against
conventional electronics. The cost of transmitting a bit of information optically halves every nine
months, as against 18 months to achieve the same cost reduction for an integrated circuit (the
latter metric is famous as Moore's law). "Because of dramatic advances in the capacity and
ubiquity of fiber-optic systems and subsystems, bandwidth will become too cheap to meter,"
predicts A. Arun Netravali, president of Lucent Technologies's Bell Laboratories in a recent issue
of Bell Labs Technical Journal.
Identical forecasts about a free resource eventually came to haunt the nuclear power industry.
And the future of broadband networking, in which a full-length feature film would be transmitted
as readily as an e-mail message, is still not a sure bet. A decade ago telecommunications
providers and media companies started preparing for the digital convergence of entertainment and
networking. Five hundred channels. Video on demand. We're still waiting. Meanwhile the
Internet, once viewed as a quaint techno sideshow for the government and schoolkids, has
transmuted into the network that ate the world. E-mails and Web sites have triumphed over Mel
Gibson and Cary Grant.
And Then There Was Light
Prospects of limitless bandwidth--the basis for
speculations about networked virtual reality
and high-definition videos--are of relatively
recent vintage. AT&T and GTE deployed the
first optical fibers in the commercial
communications network in 1977, during the
heyday of the minicomputer and the infancy of
the personal computer. A fiber consists of a
glass core and a surrounding layer called the
cladding. The core and cladding have carefully
chosen indices of refraction (a measure of the
material's ability to bend light by certain
amounts) to ensure that the photons propagating in the core are always reflected at the interface of
the cladding. The only way the light can enter and escape is through the ends of the fiber. To
understand the physics behind how a fiber works, imagine looking into a still pool of water. If you
look straight down, you see the bottom. At viewing angles close to the water, all that is perceived
is reflected light. A transmitter--either a light-emitting diode or a laser--sends electronic data that
have been converted to photons over the fiber at a wavelength of between 1,200 and 1,500
nanometers.
Today some fibers are pure enough that a light signal can travel for about 80 kilometers without
the need for amplification. But at some point the signal still needs to be boosted. The next
significant step on the road to the all-optical network came in the early 1990s, a time when the
technology made astounding advances. It was then that electronics for amplifying signals were
replaced by stretches of fiber infused with ions of the rare-earth element erbium. When these
erbium-doped fibers were zapped by a pump laser, the excited ions could revive a fading signal.
The amplifiers became much more than plumbing fixtures for light pipes. They restore a signal
without any optical-to-electronic conversion and can do so for very high speed signals sending
tens of gigabits a second. Perhaps most important, however, they can boost the power of many
wavelengths simultaneously.
This ability to channel multiple wavelengths
enabled the development of a technology that
has helped drive the frenzy of activity for
optical-networking companies in the financial
markets. Once you can boost the strength of
multiple wavelengths, the next thing you want to
do is jam as many wavelengths as possible
down a fiber, with a wavelength carrying as
much data as possible. The technology that does
this has a name--dense wavelength division
multiplexing (DWDM)--that is a paragon of
technospeak.
DWDM set off a bandwidth explosion. With the
multiplexing technology, the capacity of the fiber
expands by the number of wavelengths, each of
which can carry more data than could be
handled previously by a single fiber. Nowadays
it is possible to send 160 frequencies
simultaneously, supplying a total bandwidth of
400 gigabits a second over a single fiber. Every
major telecommunications carrier has
deployedDWDM, expanding the capacity of the
fiber that is in the ground, spending what could
be less than half of what it would cost to lay new
cable, while the equipment gets installed in a
fraction of the time it takes to dig a hole.
In the laboratory, meanwhile, experiments point
toward using most of the capacity of
fiber--dozens of individual wavelengths, each
modulated at 40 gigabits or more a second, for
effective transmission rate of a few terabits a
second. (One company, Enkido, has already
deployed commercial link containing a 40-gigabit-a-second wavelengths.) The engorgement of
fiber capacity will not stop anytime soon and could reach as high as 300 or 400 terabits a
second--and, with new technical advances, perhaps exceed the petabit barrier.
The telecommunications network, however, does not consist of links that tie together point A and
point B--switches are needed to route the digital flow to its ultimate destination. The enormous bit
conduits that now populate laboratory testbeds will flounder if the light streams are routed using
conventional electronic switches. Doing so would require a multiterabit signal to be converted into
dozens or hundreds of lower-speed electronic signals. Finally, switched signals would have to be
reconverted to photons and reaggregated into light channels that are then sent out through a
designated output fiber.
The cost and complexity of electronic switching have prompted a mad scramble to find a means
of redirecting either individual wavelengths or the entire light signal in a fiber from one pathway to
another without the optoelectronic conversion. Research teams, often inhabiting tiny startups,
fiddle with microscopic mirrors, liquid crystals and fast lasers to try to devise all-optical switches.
All-optical switching, however, will differ in fundamental ways from existing networks that switch
individual chunks of data bits, such as IP (Internet Protocol) packets. It is an easy task for the
electronics in routers or large-scale telephone switches to read on a packet the address that
denotes its destination. Photonic processors, which are at about the same stage of development
that electronics was in the 1960s, have demonstrated the ability to read a packet only in
laboratory experiments.
Optical switches heading to the marketplace hark back to earlier generations of electronic
equipment. They will switch a circuit--a wavelength or an entire fiber--from one pathway to
another, leaving the data-carrying packets in a signal untouched. An electronic signal will set the
switch in the right position so that it directs an incoming fiber--or a wavelength within that fiber--to
a given output fiber. But none of the wavelengths will be converted to electrons for processing.
Optical-circuit switching may be only be an interim step, however. As networks get faster,
communications companies may demand what could become the crowning touch for all-optical
networking, the switching of individual packets using optical processors
With the advent of optical packet switching, individual packets will still need to get read and
routed at the edges of optical networks--on local phone networks near the points where they are
sent or received. For the moment, that task will still fall to electronic routers from companies such
as Cisco Systems. Even so, the evolution of optical networking will promote changes in the way
networks are designed. Optical switching may eventually make obsolete existing lightwave
technologies based on the ubiquitous SONET (Synchronous Optical Network) communications
standard, which relies on electronics for conversion and processing of individual packets. And this
may proceed in tandem with the gradual withering away of Asynchronous Transfer Mode (ATM),
another phone company standard for packaging information.
In this new world, any type of traffic, whether voice, video or data, may travel as IP packets. A
development heralded in telecommunications for at least 20 years--the full integration of voice,
video and data services--will be complete. "It's going to be a data network, and everything else,
whether it's voice or video, will be applications traveling over that data network," says Robert W.
Lucky, a longtime observer of the telecommunications scene and director of research for the
technology development firm Telcordia.
When you ring home on Mother's Day, the call may get transmitted as IP packets that move on a
Gigabit Ethernet, a made-for-the-superhighway version of the ubiquitous local-area network
(LAN). Gigabit Ethernet would in turn ride on wavelength-multiplexed fiber. Critics of this
approach question whether such a network would provide ATM and SONET's quality of service
and their ability to reroute connections automatically when a fiber link is cut.
Life would be simpler, though. The phone network would become just one big LAN. You could
simply slot an Ethernet card into a computer, telephone or television, a far cheaper and less
time-consuming solution than installing new SONET hardware connections. Some companies are
even now preparing for the day when IP reigns. Level 3 Communications, a carrier based in
Denver, has laid an international fiber network stretching more than 20,000 miles in both the U.S.
and overseas. Although the network still relies on SONET, CEO James Q. Crowe foresees a day
when these costly legacies of the voice network will wither into nothingness. "It will be IP over
Ethernet over optics," Crowe says.
Home Light Pipes
Even if network engineers can pare down the stack of protocols that weighs heavy on today's
network, they must still contend with the need to address the "last mile" problem, getting fiber
from the curbside utility box into the TV room and home office. Some builders now lay out new
housing projects with fiber, presaging the day when households routinely get their own wavelength
connection. But cost still hangs over any discussion of fiber to the home. Until recently, advanced
optical-networking equipment, such as DWDM, was too expensive to consider for deployment
on regional phone networks. Extending the equipment into a wall panel of a split level--at perhaps
$1,500 a line--still costs more than all but a few are willing to pay. Most people have yet to take
delivery of their first megabit connection. So it remains unclear when the time will come when the
average household will need the gigabits to project themselves holographically into a neighbor's
house rather than just picking up the phone.
Dousing "Help me, Obi-Wan Kenobi" fantasies, engineers are confronting an array of nettlesome
technical problems before a seamless all-optical network can become commonplace. Take one
example: even with lightwave switching in place, one critical part of the network requires
conversion to electronics. About every 120 miles, a wavelength has to be converted back to an
electronic signal to restore the shape and timing of individual pulses within the vast train of bits that
occupy each lightwave.
Equipment suppliers also struggle mightily
with electronics envy. Component
suppliers such as JDS Uniphase labor on
methods to build modules that combine
lasers, fiber, filters and gratings (which
separate wavelengths). Building photonic
integrated circuits remains difficult.
Photons can't store charge, as the
negatively charged particles called
electrons do. So there is no such thing as
a photonic capacitor that will store zeros
and ones indefinitely. Moreover, it is
difficult to build photonic circuitry as small
as electronic integrated circuits, because
the wavelength of infrared light used in
fiber-optic lasers is about 1.5 microns,
which places limits on how small you can
make a component. Electronic circuits
reached that dimension more than a
decade ago.
The good news is that companies both
small and big are now trying to solve
problems such as signal restoration, and a
pot of venture money exists to fund them.
The field, which has taken on the same
aura that genomics now holds and
dot-coms once did, has become an
exemplar of a new, hyperventilating model
of research. Tiny development houses
proceed until they can furnish some proof
that they can make good on their promises, and then they are bought out by a Nortel, Cisco or
Lucent.
"It's a crazy world," says Alastair M. Glass, director of photonics at Lucent. "Anyone can go out
with the dumbest ideas and get funding for them, and maybe they'll be bought for big bucks. And
they've never made a product. I mean this is new. This has never happened in the past. Part of it is
because companies need people so they're buying the people. But other times they're buying the
technology because they don't have it in the house, and sometimes they don't know what they're
buying." From idea to development happens fast: a 1998 paper in Science on a "perfect mirror," a
dielectric (insulating) material that reflects light at any angle with little loss of energy, inspired the
founding of a company that wishes to create a hollow fiber whose circumference is lined with the
reflector. The fibers may increase capacity 1,000-fold, one company official claims.
Will Anybody Come?
What can be done with all this bandwidth? Lucent estimates that if the growth of networks
continues at its current pace, the world will have enough digital capacity by 2010 to give every
man, woman and child, whether in San Jose or Sri Lanka, a 100-megabit-a-second connection.
That's enough for dozens of video connections or several high-definition television programs. But
does each !Kung tribesman in the Kalahari Desert really need to download multiple copies of The
Gods Must Be Crazy?
Despite estimates of Internet traffic doubling every few months, some industry watchers are not so
sure about infinite demand for infinite bandwidth. Adventis, a Boston-based consultancy, foresees
only 15 to 20 percent of home Internet users obtaining broadband access--either cable modems
or digital subscriber lines--by 2004. Moreover, storing Web pages on a server will reduce the
burden on the network. In the U.S., according to the firm's estimate, nearly 40 percent of existing
fiber capacity will go unused in 2004, whereas in Europe almost 65 percent will stay dormant.
The notion of a capacity glut is by no means a consensus view, however.
In the end, terabit or petabit networking will probably emerge only once some as yet unforeseen
use for the bandwidth reveals itself. Like the World Wide Web, originally a project to help
particle physicists more easily share information, it may arrive on a tangent, not from a big media
company's focused attempt to repackage networked virtual reality. Vinod Khosla, a venture
capitalist with Kleiner Perkins Caufield & Byers, talks of the promise of projects that pool
together computers that may be either side by side or distributed across the globe.
Metacomputing can download Britney Spears and Fatboy Slim, or it can comb through radio
telescope data in search of extraterrestrial life. Khosla sees immense benefit in using this model of
networked computing for business, tying together machines to work on, say, the computational
fluid dynamics of a 1,000-passenger jumbo jet.
So efforts to pick through the radio emissions from billions and billions of galaxies may yield useful
clues about what on earth to do with a network pulsing a quadrillion bits a second.
raven |
