Content:
Datamation
by Alden M. Hayashi and Sarah E. Varney
What would computers be like without magnetic disk storage, graphical
user interfaces, relational databases, languages such as C++ and
FORTRAN, and operating systems like UNIX? Or, for that matter, whatever
would information technology be like today had the transistor itself not
been invented? To ask those questions is to ponder the influence of the
Big Four of IT research--Bell Laboratories, Xerox PARC, IBM's T.J.
Watson and Almaden Research Centers, and MIT.
To be sure, there are (and have been) myriad other organizations-the
Stanford Research Institute, Carnegie-Mellon University, Siemens
Nixdorf, and Digital Equipment, to name a few--that have played key
roles in advancing IT. And companies like Microsoft and labs like
Interval Research and the University of Washington's Human Interface
Technology Lab, which did not even exist when the Big Four were busy
conducting IT research decades ago, are now doing their part in the
continuing evolution of information technology.
Nevertheless, on this, the eve of the 50th anniversary of the transistor
and the dawn of a new era of network computing, we at Datamation have
taken a peek at the research currently under way at the Big Four. We
have specifically focused on technology that we think will he emerging
around IT's first fin de siecle, a time when the surging popularity of
the Internet promises major industry changes.
Holographic Storage - Crystals for data storage click here
IBM has long excelled in basic research that has helped push the
boundaries of IT. Last fall, Big Blue commenced a $32 million
holographic data storage project with an illustrious group of partners
including GTE, Kodak, Rockwell, Stanford University, and Carnegie-Mellon
University. Simply put, the technology calls for data to be stored
holographically as "pages" of bits in an optical medium such as a
crystal. Early results from experiments portend that holographic
technology could be used to store 12 times as much data at the same cost
as magnetic disk storage. Furthermore, because a laser is used for
writing and reading the data, holographic storage promises input/output
rates 10 times faster than its magnetic counterpart. "The capacity of
the technology, yes, is important. But the real feature is the potential
for enormous data rates," asserts Glenn T. Sincerbox, IBM program
manager for holographic data storage systems and technology. According
to him, data rates of 1Gbps are entirely possible.
But it is the technology's ability to perform searches in totally new
ways that makes holographic storage so compelling. Says Sincerbox: "We
can interrogate all these pages that we multiplexed into one volume
simultaneously and determine which page most closely matches our search
criteria. With rotating media such as magnetic disk storage, you're
constrained to a serial type of reading."
Sincerbox foresees the possibility of using "associative retrieval," a
pattern-recognition type of approach, for finding information. For
example, to locate information about a specific topic, a user could
interrogate the holographic crystal to find the page with the digital
pattern--or fingerprint, if you will --that most closely matches the
pattern of the information being sought. Because of such novel
datamining techniques, holographic data storage could lead to a
completely new way of representing information, assert industry gurus.
Just as IBM's development of the relational database helped transform
data into information, the company's work in holographic data storage
could turn that information into knowledge.
Yet, even if associative retrieval doesn't pan out, holographic
technology still holds promise for users with demanding data storage
requirements. For example, Sprint, which operates an enormous data
warehouse of customer information, might be interested in holographic
technology--if it proves to have the reliability of magnetic storage.
"Holographic data storage seems very' promising for now. It's on the
order-of-magnitude improvement that we need in order to justify the risk
of using a new technology," says Hector Martinez, director of business
and technical architecture for Sprint's Business Information and
Technology Solutions.
A-terabit-per-second network
In the era of network computing, the ability to store huge amounts of
data would become all the more important if you also had the
corresponding ability to transmit that data at lightning speeds. To that
end, Bell Labs demonstrated last spring a technique for transmitting
data over fiberoptic lines at a rate of one terabit per second. At that
speed, which is more than 400 times faster than the current technology,
the text of 300 years of a daily newspaper could be transmitted in a
mere second. The new fiberoptic technology uses a combination of
multiplexing techniques in which polarized light beams of slightly
different wavelengths are used to carry data simultaneously over the
same glass fibers.
Although the basic physics of the technology has been proven, several
obstacles exist. The first is that, because the light signals are
weak--in multiplexing, a beam must be split into, say, 100 different
signals-the signals need to be reamplified at intermediate points, for
example, every 50 miles. But that game can only be played for so long.
"After several such amplification processes, the signal-to-noise ratio
becomes unacceptably small, and the signals can't be received
error-free," says Andrew Ghraplyvy, distinguished member of the
technical staff at Bell Laboratories, now part of Lucent Technologies.
Thus, unless further advances are made, a terabit-per-second system will
not be suitable for long distances, that is, for transoceanic
transmission.
Furthermore, the multiplexing techniques need further refinement. "The
fibers themselves aren't the problem because they inherently have very
large bandwidth," says Chraplyvy. "The problem is in getting all the
bits on and off the fiber when you're trying to transmit terabits of
information per second."
According to Chraplyvy, the basic technology for multiplexing light at
different wavelengths is about ready for commercialization; Lucent
Technologies is offering a basic system that can handle 20Gbps. The
technique for multiplexing polarized light, though, still needs work.
"We know how to do this in the lab, but nobody's tried it in the field,"
he says.
If the technological kinks can be worked out, terabit-per-second
fiberoptic systems could accelerate the era of network computing, making
trivial the transmission and reception of large applications and huge
quantities of data. One of the strongest arguments against the Net PC,
Java, and other network-reliant entities has been that users will not
have the patience to download their word-processing applications,
spreadsheets, and other data. Terabit-per-second transmission would
nullify that argument.
The technology could also speed the acceptance of applications that
incorporate virtual-reality environments. In theory, such environments
could be generated by a powerful computer located at a centralized site
and then downloaded in a glorious gush of ones and zeros to a remote
location, instead of having to be generated at that location by another
computer that is typically less powerful. This scenario appeals to many
IS managers. "For those very highly data-intensive applications, you
wouldn't have geographic boundaries anymore," notes Steven C. Rubinow,
vice president of corporate management information systems for Fidelity
Investments.
Intelligent data replication
Meanwhile, the Xerox Palo Alto Research Center is working on improving
existing database technology to exploit today's networks. Initiated in
1991, Xerox PARC's Bayou Project is focused on delivering an
architecture for remote data retrieval that doesn't assume everyone is
linked via T-3 lines. "The world isn't well connected with high-speed
links," notes Mike Spreitzer, research manager for Bayou. In addition,
the structure of what's connected and what isn't changes over time, he
says. Thus, the goal of Bayou is to provide a database infrastructure
for handling such real-world conditions.
The Bayou technology supports any data model--relational, object, flat
files, whatever. The key is to let users access and change data remotely
in a manner that guarantees accuracy. Some current groupware
products-most notably IBM's Lotus Notes--use replication to give users
access to data. These products, however, have only rudimentary
functionality. In Notes, for example, conflict resolution between two
users seeking to update the same field simply yields access on a
first-come, first-serve basis, according to Spreitzer.
The Bayou technology deploys more sophisticated techniques for resolving
such conflicts to enable a remote client to hook up to different servers
and still receive accurate data. "The idea is to help one client see
some consistency with its own actions as it moves from one server to
another," explains Spreitzer. Specifically, a quartet of guarantees--
Read Your Writes, Monotonic Reads, Writes Follow Reads, and Monotonic
Writes--is used to ensure data integrity. "The idea is to capture some
particular writes that the next server needs to see," says Spreitzer. In
addition, users can specify how they want the system to resolve
conflicts, and this specification can be application-based and performed
by a built-in script, according to Spreitzer.
Perhaps the most notable aspect of the Bayou architecture has to do with
the notion of committed and tentative updates. Users can see two views
of the data: they can view only the results of updates produced by
confirmed writes or they can see a full view, including tentative
updates. Both views can be displayed simultaneously, says Spreitzer.
The next user interface
As the IT industry enters a new era of network computing, perhaps one of
the biggest weaknesses is the user interface. Many in the industry
assert that the graphical user interface (GUI) has simply run out of
steam. "We need to figure out some sort of network equivalent of the
graphical interface," asserts Steven Levy, technology columnist for
Newsweek.
One sure bet is that the next user interface will be more user friendly,
probably by allowing people to communicate with their computers in a way
that more closely mimics interpersonal communication. To that end, the
use of speech seems a natural progression.
Already, commercial products for speech recognition are entering the
market. Just last June, IBM introduced version 3.0 of VoiceType, which
can handle continuous speech of limited commands such as "open file
story. doe." For unconstrained recognition, for example, when a person
wants to dictate a letter, VoiceType requires that the speaker take
small pauses between words.
There are other limitations. "Right now, the system is still rather
fragile in a noisy environment," admits David Nahamoo, senior manager
for the human language technologies department at the T.J. Watson lab.
But despite its shortcomings, IBM's VoiceType is impressive for what it
promises for tomorrow. IBM hopes to achieve the field's Holy Grail--
machine recognition of speaker-independent continuous speech--within the
next five years, says Nahamoo. In preparation for that day, IBM is
working to incorporate the VoiceType technology into Merlin, the
company's next generation of OS/2.
But speech recognition is merely the first step in creating a new
voice-based computer interface. After recognizing speech, that is, being
able to identify the words in a spoken sentence, the computer must be
able to understand the meaning behind those words. To accomplish that
task, IBM, along with various other R&D centers including Bell Labs,
MIT, and Microsoft, is currently undertaking research in natural
language understanding.
More GUI miles per gallon
A voice interface may be in your future, but, for now, the GUI is here
to stay. That said, Xerox PARC has been busy pushing the limits of the
technology it helped pioneer.
Furthering research that originated at the University of Washington,
Xerox PARC has developed Magic Lens filters--arbitrarily shaped regions
of the GUI that a user can position over an on-screen application, much
as a magnifying glass might be placed over a newspaper. The lens would
perform certain functions on the selected region, and different lenses
might be combined to perform complex queries against an existing
database. Theoretically, the technology would enable even novice users
to construct complicated queries easily through a step-by-step visual
process. According to Eric Bier, Xerox PARC research manager, Magic
Lenses can be used with any kind of database as long as the data
contains correlative 2-D graphical information to enable the visual
querying.
Xerox PARC is also working on making GUI functionality more portable.
The goal is to be able to transfer a feature of an application usually
associated with a toolbar or dialog box to a movable semitransparent
sheet called a "Toolglass." Once the feature is moved, it can be used
elsewhere. "You'd learn it once, and it's good for many applications,"
says Bier.
Xerox's research notwithstanding, some industry gurus claim that the
user interface is beside the point. "The real issue is how one
represents knowledge and how one navigates through that knowledge. The
user interface is simply the mechanical device to do it," asserts Allan
Frank, chief technology officer for KPMG. The issue of knowledge
representation has certainly been brought to the fore by the burgeoning
popularity of the Internet. "There is no underlying structure to the
Web; it's simply a set of chaotic Web pages that's linked," says Frank.
"The question is: how does one put a structure on top of this?"
Web agents are multiplying
Until a way is found, users will struggle with browsers that are good
merely for wading from one Web site to another. Or, people might let
their software agents do the wading.
Software agents come in varying degrees of intelligence, the most
primitive of which perform repetitive daily tasks, such as going through
a user's email and prioritizing the different messages. Other agents are
designed to reach across a network to find specific information that a
user might want. At MIT's Media Lab, researchers have been working on
even more sophisticated agents that might, for example, automate the
purchasing function. In this scenario a buyer agent would travel around
a network to find the seller agent offering the best possible deal for a
given commodity.
MIT's agent technology, spearheaded by Professor Patti Maes, has been
spun off into a start-up company called Agents Inc., located in
Cambridge, Mass. The company's first product is Firefly, which deploys
agent technology to automate the word-of-mouth process. The initial
application is consumer oriented, a way for people to find other people
with similar tastes in entertainment-books, movies, music, etc.--through
Agents' Web site (www.firefly.com). For example, if your list of
all-time favorite movies includes many titles that are also on another
consumer's list of favorites, then chances are that the two of you would
benefit from exchanging information about other movies. "The basic
premise of this approach, called collaborative filtering, is that, if I
know you, then I can recommend things for you," says Max Metral, Agents'
chief technology officer and former graduate student of Professor Maes.
Although the initial implementation of Firefly is consumer oriented,
Metral sees possibilities for the technology to be deployed by
companies, particularly as they begin deploying intranet applications.
For example, says Metral, software agents could be used by a large
corporation that has several R&D labs to enable researchers at those
sites to learn of related work that colleagues might be doing. Through
such applications, Metral sees agent technology evolving into a type of
intelligent middleware layer that would sit on top of a database and
allow sophisticated datamining.
Agents are a huge step in the right direction, agrees Fidelitie's
Rubinow. "One of the downfalls of software," he says, "is that the world
changes so quickly that the software can't keep up with the changes. So
I think it's a great idea to have the software modify' itself."
Weaned on vacuum tubes
Historically, it has often been difficult to foresee the business
benefits of many IT innovations that research organizations like Bell
Labs, IBM, Xerox PARC, and MIT have pioneered. Indeed, half a century
ago the idea of using tiny electrical switching devices as computing
elements may have been downright silly for those weaned on vacuum tubes.
Somehow, though, transistors --as well as magnetic disk storage, the
GUI, and other innovations-have managed to become all but indispensable
in IT. Perhaps the same auspicious destiny awaits holographic data
storage, speech recognition, and software agents.
For now, though, we can only imagine how these technologies may
transform our industry as we enter a new era of network computing, which
may include talking PCs loaded with intelligent, highly intuitive user
interfaces presenting sophisticated queries via advanced software
agents. Says KPMG's Frank: "Some of the research that's going to come
out of these labs will be stuff that we won't even realize how
significant it is for another 10 years." |