octave OpenMP

[RENOO]

Hello,

I am quite new to parallele computing. I would like to use Octave on all the cores of my computer (4). Next next will be to install a cluster !!
I have installed PelicanHPC on a Virtual machine with VirtualBox. Everything went very well and it seems working properly.

Then, in Octave,  I tried to create a simple loop using parfor command or declaring my matrix with the *p special caractere,
parfor i=1:4
 A(i) = i;
end
% I start with simple things !!
Or :
 A = ones(1000*p);

but it is not working. Either parfor nor *p is know by Octave.
How should I do ? Do you have any examples?

Thank you very much for your help

[Michael Creel]

PelicanHPC provides MPITB for GNU/Octave (homepage at http://atc.ugr.es/javier-bin/mpitb) which lets you use MPI with Octave. MPI is the "message passing interface", one means of doing parallel computing. OpenMP is a different beast entirely. It is directed toward using all CPUs in a shared memory situation, like a multi-core CPU. It is not meant for distributed memory situations, such as a cluster, where MPI is the most widely used method. I don't have much experience with OpenMP. It could probably be used with Octave in a way similar to how MPITB works, but I don't recall seeing any references, You should probably check on the Octave mail list. If you're interested in MPI and Octave, then PelicanHPC is probably the easiest way to get started.

[RENOO]

Michael,

Thank you very much for your reply.
Yes it seems than learning MPI command would be very useful as it is the most widely used method. Unfortunately I need to start by installing a cluster, which I don't have yet!!. Will it be enough to create 2 virtual machines runnig Pelican on a dual-core computer ?
So far I have installed pelican on one machine and the examples provided don't run giving the following error message :" it seems that there is no lamd running on the host pelican. THis indicates that the LAM/MPI runtime environment is not operating..."
Is there a way to run the example given on all the core of one machine ?
Could you confirm me that in the case of a "real" cluster, if each node is a mutli-core machine, a parallel programme will run of every core of every nodes of the cluster ?

Thanks in advance

Renaud

[Michael Creel]

To get the examples to run on a single machine, using all available CPUs and/or cores, just open a terminal and type "lamboot". To use more than one machine, follow the instructions in the Tutorial. Also, you should check the documentation for LAM/MPI or Open MPI to get a better understanding of how it works. Things are a little complicated when you're beginning with MPI...

[RENOO]

Michael,

Thank you very much very your precis answer.
I wanted to start slowly and using a virtual machine running Pelican before installing a cluster, given my knowledge in parallel computing.
I have used Virtualbox to install Pelican on a host fedora C10.
Unfortunately, it seems that virtualbox dont not allow the use of all the core of the physic machines.
I guess I will find a dedicated PC to install Pelican and come back later on this project.

Best Regard