[Beowulf] newbie

[Tomislav Maric]

Thank you all very much for the advice, knowing that I can find help and
advice here is really great, thank you! I apologize in advance for many
newbish questions, and the delay in answering, I've had to switch to
Thunderbird because Evolution gave me problems and gmx.com app didn't
have a switch for plain text messages.

@John Hearns

I'll take Your advice and try and focus on the goal: setting up the
clusters (at the lab and at home) and learning what I need as I go. My
mentor is one of the starting developers of OpenFOAM and he maintains
the sources on the SVN repo. He and the other developers have put
OpenMPI and tons of other programs in the -dev version of the toolkit,
so I've already ran parallel simulations on my dual core laptop. I've
used this command:

foamJob mpirun -np 2 nameOfTheSolver -case caseDirName -parallel

and I've heard from a colleague of mine who's a PhD student in Zagreb
(also doing work on OF) that he ran many simulations on the NOW in the
lab. I've had experience postprocessing data from around 100 CPUs (my
mentor ran a case on some cluster somewhere). Right now the network in
the lab is down: I need to mirror the installation from the head node to
others (without OF, of course), so I'll do as You have suggested. Only,
I thought there was more to be done on such a network to beowulferize
it, besides having OpenFOAM and OpenMPI working on it?

Thank You for the job advice, to be honest, I the thought has  crossed
my mind that it would be great to have HPC experience when applying for
a job (i love computational physics and Linux and the whole Open Source
have gotten under my skin, so I would LOVE to work in this area), but
I'm doing this because it really interests me (enough to spend my own
money and time), and also for another reason:

a boxed cluster was bought a year and a half ago for the lab, and all we
have from it are problems. If I assemble a small beowulf at home, on my
own, I can show the "proper authorities" at the University the right way
to do things.

Thank You!


@Andrew M.A. Carter

This is exactly the kind of approach description I was looking for!
Thank you! Now I have a general outline of my work. We have some old
processors (single core 2GHz - not bad) and motherboards, so this will
be the start for me.

[I don't have so much money... the 500 euros is about the limit, and
with the processors and motherboards from the old network, I hope it
will do]

>"I have an
> assignment on computer clusters: can you tell me all about
> Beowulfs and what they are good for

I wish only that I could listen to such courses... :) No student credits
on this job for me, besides my own satisfaction. :)

@Joshua mora acosta

Thank you for the advice regarding the questions. :)

As for the finances: me and my friend (the above mentioned PhD student)
are using Open Source in all occasions, and the resistance in the
Faculty is really high. The only way I could get some funding is to make
something, prove that it works, prove that it's cheaper, recyclable,
robust and 999999999 times better then the "closet computer" that
we can't use because of many complications we've had in dealing with the
company that sold it. They sold us the machine (I'm just a student, so I
don't know how much it costs) and they don't have Linux admin in the
company. ALL of our simulation software is GNU and runs on Linux (it
runs on other oses also, but... I'll be polite here and just stop
writing).

Since I have some hardware left and some money from the scholarship I
think I'll try and do something at home in parallel to setting up the
network at the lab. After all, I'm getting my Master degree in a year,
and leaving the University - if I'll work on my own time, without
getting paid (in any way), might as well have the fruit of my work in my
apartment. :))

@Prentice Bisbal || Greg Lindahl || John Hearns || anyone

Why Intel compiler? OpenFOAM is compiled with gcc. My mentor instructed
me to compile it with the same gcc - we communicate via e-mail because
he's spread all over the world, so I'm keeping the differences in our
builds as small as possible.


Cheers,
Tomislav