Roy Wilson wrote:
>
> Niko,
> I'm just curious: what sort of simulation is involved?
> Roy Wilson
> [EMAIL PROTECTED]
>
> --On Thursday, February 24, 2000, 9:10 AM -0500 Niko Nicopoulos
> <[EMAIL PROTECTED]> wrote:r
>
> > I have set up a cluster of 16 dual processor boxes for Monte Carlo
> > simulation at my company, each running SMP-Linux. I use PVM both within
> > and across boxes to run parallel applications. I experimented a while ago
> > with using threads within each box, and PVM or MPI for message passing
> > across boxes, but I found out that neither of these pieces of
> > software work well with threads.
> > After I inquired, I was told by people who were involved in supporting
> > these two packages, that they knew of no reason why these packages would
> > be incompatible with threads, but they are. I was also told that there
> > was some work underway to address this, but apparently nothing has been
> > done.
> >
> > By the way, I am still using an Oooold kernel (2.0.36) compiled with SMP,
> > because the last official (albeit development) SMP kernel
> > I tried (2.1.15 or so) gave me trouble.
> > My PVM applications would run only if compiled under the old kernel.
> > Mind-boggling!! Has anybody looked into this?
> >
> > Nikos Nicopoulos
> > -
> > Linux SMP list: FIRST see FAQ at
> > http://www.irisa.fr/prive/dmentre/smp-howto/ To Unsubscribe: send
> > "unsubscribe linux-smp" to [EMAIL PROTECTED]
>
> -
> Linux SMP list: FIRST see FAQ at http://www.irisa.fr/prive/dmentre/smp-howto/
> To Unsubscribe: send "unsubscribe linux-smp" to [EMAIL PROTECTED]
Well, the basic technique is to have a set of n-dimensional coupled
differential equations. Then you just step out until some error margin
is reached.
On the way you write to a file - you need plenty of disk space.
When you have the data you need, you start to analyse them - plotting is
an intuitive method - but will often confuse you.
To plot I use explorer from SGI.
The diff. eq. can change depending on what system you wish to describe -
basicly you look them up in a physics book...or you remember them, if
you have had a physics course.
Ofcourse you can use the same technique to model fluctuations in stocks,
chemical reactions or even friendship relations.
You need some rules, which can be expressed as formulas.
The parallel stuff comes from the set of diff. eq. This set can be
expressed as matrixes, and matrix algebra is ideal to do is a parallel
way. Ofcourse there are plenty of other calculations that are more
efficient doing in parallel - the point is just that some are not: hence
it is stupid using time to do them parallel.
I have a friend who is very interested in the monte carlo simulations.
At Niels Bohr institute we have a cluster - could we see the source for
the simulation ?
nino
--
Venlig hilsen / Best Regards
Bernino Lind
_________________________________
Bernino Lind - iSolve ApS
R�mersgade 7 kld
DK-1362 Copenhagen K
Denmark
+45 33 32 59 78
-
Linux SMP list: FIRST see FAQ at http://www.irisa.fr/prive/dmentre/smp-howto/
To Unsubscribe: send "unsubscribe linux-smp" to [EMAIL PROTECTED]
