• Presentation
• Register
• Program
• Organizing Comittee
• Listing of registered people
• Translators team

> quien va a traducir ?
QuArK	yo ayudo
QuArK	y damage y tu?
> yo no se :(
> tu tambien venga
Mukak	jajajaja
Mukak	me encanta esta organizacion
Mukak	"on the fly"
-> *one2* op #redes quark
damage	ups
> ayudas tu tambien ?
QuArK	is hard
QuArK	?
Mukak	hmmm
Mukak	pero como hacen en grupo ?
Mukak	es medio complicado .....
QuArK	bueno nos dividimos los parrafos
QuArK	traducimos de a 6 lineas cada uno
QuArK	example
QuArK	yo traducire esto
damage	ya
damage	al pv
damage	hello today we have with us to David Santo Orcero, another
veteran of this congress. We can say many things of him, but better we let his
char speaks it of its merits. It goes to speak to us of the project Piles up
surely and that of other interesting things. Thanks to David to prepare this
presentation. Thanks also to all you to go.
QuArK	dame op
QuArK	bueno cordimenme que traduzco
damage	The year last in this same congress I commented the
architecture Piles up was only an idea fruit of brainstorming between Robert
Holy and I for whom it does not know that the subject goes, it is a
burglaryable architecture of grid that allows that software as much as the
hardware is free.
damage	(more information http://umeet.uninet.edu/umeet2003/english/ta
lks/20031222.3.es.html)
damage	Create some possible problems and its solutions for an
architecture of the future the idea is not a patch to Linux, but to raise
subjects that are implemented in 5-10 years and that allow that Windows did
not go ahead to us. We continued implementing UNIX; and eye, that Microsoft
is with new paradigms
damage	of computation. The idea of Piles up is to play with the new
paradigms. I recommend the reading of the previous year before this
conference. The first point of that we are going to speak is of the
migration. The original idea was to use the algorithm evaluation of the
optimal conditions of of migration of OpenMosix, developed by professor
Amnon Barak for the Mosix project.
damage	This algorithm, as all we remembered, has demonstrated that
for an infinite time the advantage of cluster is similar to the one of an
optimal algorithm of finite window - that is to say, an algorithm in which
we know the future, but not _ everything _ the future. On the theory it was
very well, so I began with the simulations.
damage	The form to simulate this is easy. Evidently I am not going
to move machines in a simulation, but I can make script to raise and to go
down to interfaces, simulating therefore the behavior facing the operating
system of a displacement of grid within a space of greater other grid. I am
speaking of _ physical displacement _ that is to say, that somebody with a
PDA or a Wearable Apila happens from the hall to the kitchen.
damage	With which constantly there are machines that enter and leave
the network visibility of the machines that produce the processes that migrate
outwards. The processes were blind curls in C. (in the simulation, the
experiment assumes that in the system It piles up will be common applications)
In principle, was hardly to verify that something worked that supposed that it
was going to work.
damage	Nevertheless, the result was disastrous. Still making trap -
the migration of return of the processes forced before disconnecting the node
-, the advantage of the resources was very below the theoretician. My first
supposition is that they were few nodes to be able to interpolate results.
Duplicating the number of nodes, unless I needed one more a greater machine to
make the simulation and that this fué much more slow, I observed a result
still worse.
damage	Evidently it was ignoring something. The only logic is that
there was some implicit condition in the demonstration of the algorithm that
it had violated in the simulation. Return to the article, analysis of which
I am doing, and after a time comes the illumination. (what it seems a short
process of inference took more time of the one than it seems); -)
damage	The algorithm of calculation of the optimization to migrate a
process or another one in OpenMosix leaves from the assumption that,
although the nodes have something of trustworthy information on part of
cluster, the size and the characteristics of cluster remain more or less
constant in the time, or change little in the time. In surroundings of grid
as the proposed one in Piles up, this does not have sense. It is necessary
to change something.
#redes
damage	Are put machines to feed the computacional power on Piles up,
or producers that in a while in individual do not have load. The producers
of tasks to each other have a stability that causes that they do not
present/display a great variability; that is to say, which they do not move
to each other for the surroundings that surround a particular group by
consumers can have discharges, losses or sporadic movements but must not
have clay variations in aceso
damage	(sorry, I have realized that last the four lines I have
interchanged the words producing and consuming the problems of the direct
one; -) ) that is to say, I can be left the PDA piles up upon the table, and
to go to take coffee but would have to be sporadic - or I will remain
without PDA -. the group of the producers can let see consumers at any time,
and see new consumers. This corresponds to the case that the human who takes
the devices Piles up abo
damage	The idea, in 2D and extrasimplified is:
damage	<irbis> t1:
damage	<irbis> CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC
damage	<irbis> PP
damage	<irbis> t2:
damage	<irbis> CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC
damage	<irbis>  PP
damage	<irbis> t3:
damage	<irbis> CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC
damage	<irbis>    PP
damage	I believe that you understand the idea. The jump to see itself
not see itself in normal conditions is not steep. This means that a consumer
sees as he loses the power of signal with the producer, until he lets see it.
As we solve the problem? Let us suppose that we assigned slots to each
processor Piles up, according to its power of calculation. The normal models,
with the small power, have as soon as slot.
damage	(we are speaking of systems Piles up with power of a
microcontroller) more powerful Processors will be able to have more of slot.
A producing processor that actue as will have more processes than slots. A
consuming processor that actue as will have more slots than processes. A
processor with identical number of slots that processes is taken advantage
of to total capacity, reason why does not interest to us to treat now on him.
damage	We suppose that the operating system of a processor Piles up
can know the power the connection signal and its variability in a last time
(under the speed, the translation is at full speed working) with any other
processor Piles up at any time of the surroundings. We defined migrations of
two types. The first type of migration is of producer to consumer. Whenever
there is a visible producer and a consumer with a power of signal up to a
threshold the mi
damage	The second type of migration is of consumer to consumer When
a consumer who executes a process generated in a producer sees like loss the
power of connection with this producer underneath the soft threshold, the
consuming speech with the producer. The producer verifies the conditions of
migration according to the previous algorithm, and informs to the consumer
where it must send the task.
damage	Evidently that negative events can happen - interferences, lost
of connection -, but in normal conditions it works at least. Now I have left
to try to formalize the algorithm mathematically: -) With respect to the
processor It piles up From year principle I have been speaking with Robert
Holy by the subject of the processor. It has sent enough documentation to me
to evaluate if a chip can be programmed that has like native language the CLI
of .net
damage	I am not going to evaluate if he is interesting politically or
technologically. Hardly we are going to go to the objectives of the project.
The good news: it is possible. In fact, there is people who already have done
it. the bad news: probably the design of a processor that interprets CLI
natively is so complex that it a fan in his house cannot do, reason why is not
worth at the present time for the free hardware.
damage	Personally, if transistors exceed to us, I prefer to put more
registries. The Java processor also is a bad experience on the matter. I
already take to a pile of time giving *tabarra. Passage to the debate. But
before comentaros that there is a list of mail on Piles up in
apila@orcero.org
damage	I wait for your commentaries on the subject in which we are
working, and that you animeis to work in the subject: -) Questions? good, to
the next one! thanks for your attention.
* damage sets mode: -m

Generated by irclog2html.pl by Jeff Waugh - find it at freshmeat.net!