Reply to The New York Times

Connect, They Say, Only Connect, January 25, 2003 By EMILY EAKIN

 

The Devil is in the [Switching] Details

By Borut Prah

Connect, only connect...  To what?  Optical fiber network?

Can I really connect a beam of light from me to you? 

 

The answer and the problem lies right in front of
our noses-- the Telegraph Hill.  It was a message relay station. 

Here was one of the early hubs of a network that used

light to carry information.  The problem was not the

speed of transmission, say, from Fort Point to Telegraph

Hill, but what happened to the photons after Telegraph Hill

received them. A mirror on Telegraph Hill, if turned exactly

in the right direction towards the destination would relay

the message instantly.  But since this is hardly possible,

the Telegraph Hill operator wrote the message down and

relayed it.  


Two years ago, before the stocks of Cisco, Nortel,
Lucent, Juniper Networks, Enkindo, JDS Uniphase,
Nanovation, Agilent, Corning, Chorum Technologies,
Calient Networks, Optical Micro Machines, et al., took
the big hit or evaporated altogether, the signal of the
trouble with switching light was well publicized. The
Scientific American January 2001 issue quoted Alastair
M. Glass, director of photonics at Lucent: "It's a crazy
world.  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".  And they

still have not...

The network can routinely shuttle trillions of bits
in a second between hubs.  But switching it in the hub,

from a source node to a destination node is just as difficult

as pointing a mirror on the Telegraph Hill in the right

direction. 

 

Airlines have discovered this problem with passenger

traffic in a hub system. A jet will take you from San Francisco

to Denver hub in three hours.  But you want to go to Atlanta

and the wait in Denver is five hours.  Not only that, you

must walk a good distance to another gate.  Same problem

exists with switching light.

The electrons sent from your PC must become photons

before they can travel over fiber networks.  That is easy.  The
problem arises when your photons reach a hub and must be

switched to the specific address that you want them to reach

among the millions of  addresses.  But photons, unlike electrons, 

cannot be stored without some very expensive tricks. 

A message carried by light must be switched by mirrors

instantly, or it is lost.

Billions of dollars have been spent, lost, and even
wasted in the effort to switch information packets with
light.  Tiny mirrors must be constructed on microchips
that switch light beams from million of sources to
millions of destinations in millionths of a second.  But
this is not an easy matter and it is far from solved.

Thus, interim solutions have been constructed which
convert photons back to electrons, then store them in
the memory chips, such as you have in your PC, from
where they can be switched back to the right optic fiber
to reach destination.  The result is a delay, analogous to

waiting for a connecting flight.  The actual process is

even more complicated since the internet protocol breaks

up information into packets that are switched many

times before the message reaches the destination.

Typically a message travels from a sending node to a

router hub to another router hub and so on until it 

arrives to the recipient node.  Since hubs do not yet switch

with mirrors but with electronics, the phenomenal carrying

capacity of the fiber is only attainable between hubs. 

 

Only heavy business users have applications that can justify 

hub to hub links.  The millions and millions of internet users

have little reason to connect if it takes hours, or maybe

even days to download a movie through electronic switches.

 

Until the photonic switches come of age, over 90% of fiber

optics channels will remain dark.  And so will the prospects

for the return on investment made into the network trunk lines. 

They are all dressed up but nowhere to go.  Yet.

 

PS July 2003: CORVIS (NASDQ: CORV) now has the answer.