MATLAB: Is it possible to use Parallel Computing Toolbox 4.1 (R2009a) in an InfiniBand Switched Network

Parallel Computing Toolbox

I would like to use Parallel Computing Toolbox and MATLAB Distributed Computing Server in an InfiniBand network.

Best Answer

Parallel Computing Toolbox 4.1 (R2009a) does not have built-in support for InfiniBand. However, the toolboxes provide all the necessary hooks to take advantage of it.

For this, the user will need to provide their own custom build of MPI that supports InfiniBand. Instructions on how to use a different built of MPI can be found in the attached document under the subtopic "Using a Different MPI Build on UNIX Operating Systems".

Related Solutions

MATLAB: Workaround for Parallel Computing with LSTM

Hi Aditya,

Using multiple workers are not yet supported for LSTMs or recurrent networks. Unfortunately, there is no workaround for training recurrent neural networks like LSTM in parallel. The developers are aware of this behaviour and will try to provide a workaround in future releases.

Hope this helps.

MATLAB: How to use the Parallel Computing Toolbox 4.1 (R2009a) to find the singular value decomposition of a large matrix

To determine the singular value decomposition (svd) of a large matrix in the Parallel Computing Toolbox 4.1 (R2009a) without using pmode, create a MATLAB function file that creates a codistributed matrix and finds the svd of the matrix.

The concept is illustrated using an example given below:

function y = thisJob(n)
x = rand(n,codistributor());
y = svd(x);

Execute the following commands to run the above job on the worker machines:

main.m

An example of math with a codistributed matrix in non-pmode

% Find the available parallel computing resources
sched = findResource('scheduler','type','local');
% Construct a parallel job object using the resources
pjob = createParallelJob(sched);
% Set the dimension of the random matrix
a = 16;
% Set the file thisJob.m so that the workers can access it.
pjob.FileDependencies={'thisJob.m'}
% Add the task to the job.
createTask(pjob,@thisJob, 1, {a});
% Run the job.
submit(pjob);
% Retrieve job results.
waitForState(pjob);
out = getAllOutputArguments(pjob);
% Display the random matrices.
%  Note:  the output values in each worker will all be the same, they
%  represent the singular values of the matrix.
out{1}
% Destroy the job.
destroy(pjob);

Best Answer

Related Solutions

MATLAB: Workaround for Parallel Computing with LSTM

MATLAB: How to use the Parallel Computing Toolbox 4.1 (R2009a) to find the singular value decomposition of a large matrix

Related Question