| |
I.
Network technologies: Metcalfe's magic square
Robert Metcalfe, the creator of the Ethernet
protocol and founder of 3COM, has formulated
an empiric law to measure the usefulness of
a network:
Network
usefulness= k * N²
|
where
N stands for the number of network nodes and
k for a multiplying factor.
In
other words, the usefulness of a network is
proportional to the square of the number of
nodes that are connected. Let's take the example
of the telephone. Starting with two users,
there is a network, but it’s not very
useful. Once most residents in a city are
connected, the usefulness of the network grows
larger in proportion. Today, virtually all
people living in the developed world are connected,
so the telephone network has created a "global
village."
Diagram
1: Metcalfe's law
II.
Internet : the premature revolution
Let's
now consider the case of the Internet. In
1993, 2.5 millions computers were connected.
In 1997, they were 25 millions. Using Metcalfe’s
formula, the usefulness of the Internet was
roughly multiplied by 100 over four years.
This mathematical assertion can easily be
verified in our everyday life, which has been
dramatically changed by the Internet. Writing
and receiving e-mails has become routine.
We are accustomed to getting news, entertainment
and many other forms of information through
the Web. All this led us to think the Web
was a revolution. Even the financial markets
eventually adopted this belief, resulting
in the "Internet bubble" of the
late 1990s.
Then,
in 2000, the bubble burst.
Why ? Because then, Metcalfe's law applied
only partially to the Internet. True, millions
of computers were connected, but the Web still
had a centralized architecture, where most
of the electronic exchanges took place between
Web users and Internet sites or portals. Within
that hierarchy, there was little direct interaction
between individual Web users.
III.
Cooperative computing: the (really) big leap
forward
The development of cooperative computing may
signal the beginning of his long-awaited revolution.
The idea is simple: take advantage of Internet
connections to bring together computer resources
disseminated all over the world. The phenomenon
has gathered unprecedented momentum over the
past two years due to the availability of continuous
broadband connections. We are gradually approaching
the point where every individual computer on
Earth can interact with all others, bringing
Metcalfe's law's to its conclusion.
IV.
Four steps towards cooperative computing
Cooperative
computing has developed through four application
areas, paving the way to distributed intelligence.
- Sharing
information: peer to peer
P2P
has become increasingly popular during the
past 5 years. Napster was a prime mover. Its
system allowed millions of music fans to exchange
songs for free by transfering sound files
(mostly MP3) from one computer to another.
Feeling the threat, music publishers sued
Napster, arguing that it encouraged piracy.
Napster's weak point was that its technology
used a centralized server, so the company
could not deny knowing its users were engaged
in illegal traffic. In the end, Napster was
closed down. Yet that did not stop file sharing.
Instead, it fostered peer to peer systems.Programmers
developed truly decentralized applications
and protocols, including Gnutella, Kazaa,
eDonkey and Morpheus. The companies that produce
these software programs cannot be easily accused
of piracy, because with no central server,
they cannot be held responsible for how people
use their technology.
Beyond legal disputes, one thing is certain:
file sharing can be used to make information
available as it has never been. With peer
to peer, there is no need to publish information.
If it is on the shared folder of a connected
computer, it can be instantly retrieved by
any other computer on the network. (1).
- Sharing
computer resources: grid computing
Some
applications require such high levels of computing
power that they require the use of expensive
supercomputers. These applications include
"big science" (astronomy and physics),
finance and biochemestry. Industry also needs
more and more computing power as it shifts
from real world experiments to simulation,
whether for designing aircraft or for assessing
a car's safety through virtual crash tests.
All these applications require intensive computer
resources. This demand can already be satisfied
(or soon will) through cooperative computing.Instead
of allocating the entire set of calculations
to a single computer, why not split it into
millions of pieces and ask many computers
all over the world to do their share of the
work ? All it takes is a simple installation
of a small program that will process the data
sent by a central server when any given computer
is idle. Then the PC sends back the results
to the server, which combines the millions
of partial results to reconstruct an overall
solution. This intelligent technique, known
as grid computing, can be used in various
areas:
-
Science
Grid
computing is a viable alternative for carrying
out huge scientific programs without huge budgets.
One of the first examples of this was the SETI@home
initiative. (Search for ExtraTerrestrial Intelligence).
The goal of this research is to analyze signals
received from space in search of a logical sequence
that could not be the result of a natural or
human phenomenon. Many frequences, coming from
many directions, could contain such a signal,
resulting in a tremendous workload for computers.
That is why the calculations were shared among
several millions of computers owned by individual
volonteers around the globe. According to SETI@home
officials, there are currently three million
PC computers participating in the program, which
have the processing power of a 15 teraflop machine
(a computer capable of 15 000 billion mathematical
operations per second). Yet the total cost of
the program does not exceed $500,000 By comparison,
IBM's ASCI White, the most powerful computer
today, has a capacity of only 12 teraflops and
costs $110 million.
-
Medical research
Will
SETI@home discover an extraterrestrial signal
? Maybe or maybe not. But other programs make
the most of grid computing, with more direct
benefits. Stanford University is managing a
program aimed at studying genoms and protein
synthesis. The program consists of two parts,
Folding@home
and Genome@home.
Since it started in early 2001, more than 20,000
CPU years have been donated by volunteers. (CPU
years are similar to man-years for a human-based
project.) This effort made it possible to simulate
over six million new proteic sequences. It helps
scientists better understand the way proteins
fold in nature (how they assemble), which is
a key but still mysterious biological phenomenon.
-
Business
Many
business sectors require enormous computer power,
including finance, car manufacturing and biotechnology.
Buying supercomputers requires heavy investment
that can be avoided by setting up computer grids.
Take the example of a bank. In order to carry
out its complex financial operations, it will
be able to use the idle time of computers on
its Local Area Network (LAN). This solution
has many advantages. First, it is relatively
cheap. Second, it is scalable. If the bank needs
more computing power, it will only have to tighten
its grid by adding more computers to it. If
the needs vary in time, the bank will be able
to outsource its supply of computer time. A
subcontracter will sell it the exact number
of CPU hours required, just as the electrical
utility invoices for the number of kWh it consumes.
Some companies are already providing this kind
of service, like Gridsystems,
through its solution "Innergrid".
- Sharing
knowledge: expert networks
Knowledge
management is also a part of cooperative computing.
It addresses the fact that much of the knowledge
capital of a company is implicit and remains
in the brains of its current employees. This
goes far beyond official qualifications, titles
and site locations. In order to make this
"knowledge treasure" available to
the entire company, so-called Employee Knowledge
Networks (EKN) offer a way to connect an employee
trying to solve a problem with other employees
who have the answer or parts of the answer.
To do so, employees must be encouraged to
feed the company's knowledge databases, to
participate in discussion groups over the
intranet, and to identify themselves as "experts"
on given subjects. Then a series of technical
tools, including e-mail, instant messaging
and virtual workspaces makes them available
to answer questions. Askme corporation provides
a good overview of the benefits of an EKN
on its homepage.
- Sharing
brains: distributed intelligence
The
most sophisticated aspect of cooperative computing
is distributed intelligence. The idea is to
capitalize on individual behaviors to create
a superior collective intelligence. This could
be compared to the way ants organize their
colonies. Each individual ant has a very limited
intelligence, but by assembling thousands
of tiny intelligent behaviors, the colony
is organized as a complex and effective structure.
Several examples of distributed intelligence
are already in evidence, such as the fight
against spam. Cloudmark
sells a clever solution called Spamnet to
unite the efforts of Internet users around
the globe.Every time a member of the community
receives a spam e-mail, the user can label
it as such. This information is instantly
transmitted to all network members. so they
will all be protected from the message. When
the next person receives it, it will be immediately
quarantined in the Spam folder of the mailbox.
Another initiative is Worldwide
Lexicon, a program aiming at creating
an automatic translation system for rare languages.
It relies on several thousands volunteer translators
in various countries who agree to translate
short sentences or keywords during their spare
time.
Diagram2:
The four steps towards cooperative computing
In summary, cooperative computing may be the
path to a new era of the information society.
Concepts like sharing, community and participation
play the leading role. Just as we pay no attention
to the generation of the electricity we consume,
individuals and businesses will have immense
computing power to command in the near future,
originating from an infinite number of tiny
resources. This should enable intelligent
processes that are still difficult to conceive.
|
|
