From: christophe delriviere (darkmichet@bigfoot.com)
Date: Tue Nov 17 1998 - 18:45:39 MST
Mike Linksvayer wrote:
> Spike Jones wrote:
> > primes may not turn everyone on. what if distributed task
> background
> > computing became popular. what else could you think of to use all
> > that idle computer time? spike
>
> Ever since reading Vernor Vinge's "True Names" I've thought a
> distributed
> artificial intelligence would be one of the neatest things one could
> do
> with distributed computing (the "Mailman" from the aforementioned
> story
> was a distributed AI that got its name because it answered questions
> slowly, as if by post).
>
> --
> See From: and Organization: above. Call +1 415 553 6408 for
> assistance.
>
> (first born
> INTP)
Several months ago, i had written a very little essay about that (in
french).
i don't know too much of the state of the technology right now, or if my
ideas are good ;)
i've no credentials after all ;) so i would be glad if somebody tell me
that i'm completely wrong ;) ..
having not too much time to translate it, i've used the Altavista quick
translation, so be prepared ;)
**
Tracks for an automated approach of the distribution and sharing of the
scientific computations requiring the rough force.
**
Currently, certain groups of people use a highly parallel approach to
discover the long keys of protection of the
systems of encryption. **time-out** the work, use primarily the force
rough be share in a community of user have
for certain access with some vast resource like some network of
university or some large company and for some
other have access with some resource more modest like some small
computer network in PMEs.
Responsible people or having access to an information processing system
are often characterized by a same
common denominator. Their systems are not used with a maximum output.
Often the machines under-are
exploited, during many times of pause, during the night, or while
nongreedy applications in power processor are
carried out. To fill the objectives of their organizations, the outputs
of their networks cannot be optimized in an
economic way or they consider that these outputs do not need to be
increased, the awaited profits being for
example tiny.
A current very interesting example of a typical step of automation and
massive parallelization of a problem
requiring the rough force can be found on " ' SETI@HOME homepage " with
the following address:
http://setiathome.ssl.berkeley.edu / the SETI decided, via Internet to
make it possible to the users of PC to spend
time when they do not use it with profit for his calculations. This
approach enables him to achieve goals
impossible to realize without investing enormous means compared to its
economic possibilities.
More and more of electronic apparatuses comprise processors which they
are rudimentary or very elaborate,
good markets or very expensive. This tendency will make only increase
and all the types of apparatuses tend to
incorporate this type of component, even if they do not need any in
their minimal form. The price of the
processors makes only fall, it thus becomes less and less dramatic to
integrate a too powerful processor for the
application required by the apparatus in which it is built-in.
The next standards planned for the Internet will be characterized by
higher flows of transfers of data but also by a
number of almost inexhaustible IP addresses. One will be able to thus
give an Internet address to any object
connected with the network. One will be able to give an address to the
cars, the watches, GSMs, televisions, the
pacemakers, and their components, etc. **time-out** it type of approach
can allow in particular the control
remote of these objects by their owner or by of system of management or
still the implementation of system of
diagnosis of breakdown automate allow the setting in functioning of
correction software for avoid with defect
appear and/or of make intervene a system of repair as soon as possible.
All these processors will be still characterized by this same problem of
under use.
The number of simulations, algorithmic calculations scientific requiring
a rough force of calculation is
potentially astronomical. **time-out** that they himself be of program
require a execution of nature more sequential, a parallelization more
significant can generate a overhead too high, or some program require a
execution of nature more parallel, it be necessary seek the means of the
executer the more effectively and the more quickly with the less some
resource possible. To reduce the total costs to some extent.
A very significant step to accelerate their executions would comprise
the following stage:
Establishment of a committee on a world level for the management of a
data base counting the various problems
of scientific nature which one can solve by methods of massive
parallelization. **time-out** the committee
himself occupy also to carry out ŕ a classification by command
preferably of these problem, define the tactics of
distribution automatic of code de computer with processor unoccupied and
to count the result and them put
accessible with public.
For very many and increasing class of owners of electronic apparatuses,
the knowledge that their acquisition
contributes to the advance of search constitutes one more been worth
obvious of those. It will thus not balk for
their majority to invest a negligible overcost in their apparatus to
allow the installation of such a system.
The procedure of the committee could be described not the following
elements:
**time-out** creation of a system of payment as for method of
distribution of code executable with processor
door frame, with management of different priority of experiment which
must be distribute according to certain
criterion define by some function of probability distribution de
probabilité.
Reception of proposals for a search requiring of the numerical means of
scientific organizations or the public or
of other organizations.
The criterion of dormant state of the processor, can be defined by the
manufacturer or the user according to his
characteristics.
The processor in dormant state sends a package to announce its state to
the server of the committee. The server sends to him then according to
its capacities of calculation an executable code expressed preferably in
the most universal possible language, like Java for example.
When the processor is not any more in dormant state, it sends its
results to the server. Results being in general a standardized
description of the place where it finished its execution. In this way
work can be resumed by another processor in dormant state. When a
processor finished the total execution of a code, it returns the final
results and can then receive a new code.
One of the major side effects of such an approach would be to be able to
check that it is not possible to make emerge an " intelligent " system
of a gigantic network of neurons answering certain criteria of
implementation. Indeed one could carry out this network of neurons on
all the processors in waking state
Even if this system is not really " intelligent " it would constitute in
any event a fabulous tool for search having in particular all the
scientific resources of the Internet.
An article of the edition of May 98 of the Byte magazine treated search
in progress and currently available softwares being able to make more
easily realizable relativemment, from a technical point of view, such a
aplication. It is about " Millipede software turns even the smallest
network into has virtual parallel machine " of Tania Hershman.
Some sites of references are quoted:
Brazos Project: http://wwwece.rice.edu/~espeight/brazos.html
Project Calypso: http://www.cs.nyu.edu/milan/calypso/index.html
Millipede Project: http://www.cs.technion.ac.il/Labs/Millipede /
TreadMarks Project: http://www.cs.rice.edu /
ahh... too bad this free translation... does somebody know about a
better free translation system ;)???
well i send it anyway...
delriviere
christophe
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 14:49:48 MST