[comp.arch] Electro-Optic Busses

baker@necssd.NEC.COM (Larry Baker) (08/11/90)

[]

Awhile back someone asked about an anonymous company creating an optical
circut, and promising to deliver a finished application by 2000.

AT&T demonstrated a *working* optical switching matrix at SuperComm in
Atlanta, April of this year.  They had two working prototypes at the show,
with Bell Labs designers there pointing out salient features.  I forget
the details, but it was all written up in one of the IEEE rags around March
1990.  I may have a copy of the article around my office; I'll look.  Also,
about a month before they announced the working switching matrix they
announced a working adder built out of the same basic technology.  One of
the problems, they said, was that it could operate at speeds beyond their
ability to test it, and they felt that optical computing at speeds that
truly stretch the limits of the design materials would be a long time
coming, due mainly to that problem.

AT&T claims that they intend to deliver a working switching product
using the optical matrix as its core switching fabaric, in 5 years.  This
is extremely audacious and probably unrealistic, but it just may be
possible.

One of the problems with the prototypes, however, was that it used an
electronically pulsed laser to change the matrix connectivity, so the
actual *switching* speed of the matrix fabaric cannot exceed the speed
of the (electronic) control logic.  Also, signaling for switching at such
high bandwidth (gigabits/sec/channel) is effectively impossible, because
electronics have to interpret it and respond with changes to the matrix,
a double trip through the electro-optic bridge.  My guess is that they'll
use it for some kind of monster or broadband  switch, with conventional
64kbps and ISDN/BISDN channels multiplexed through the matrix and a
stored-program or hardwired computer telling it what to do.

But, if you think about it, once very basic signaling and matrix management
can be handled directly within the matrix fabaric (no electro-optic
bridge), what you have is a hell of a fast multichannel bus.  Think of some
of the supercomputing applications that could take advantage of terabit
connectivity, switched at bus speeds.  It brings whole new meaning to
parallel computing....

Cheers,

LEB