Prev: find user and org info for current install?
Next: R2008a Windows Installation Woes
From: Jaime Laviada on 7 May 2008 05:21
I am doing some preeliminaries test in a minicluster with
the distributed computing engine toolbox from matlab 2008a.

My minicluster has three dual core 2 with WinXP 32bits and
one dual core 2 with WinXP 64 bits. I create two workers per
machine.

The problems arises when I create the pool with all the workers:

-------------------------
matlabpool('OPEN', 'jobmanagerconfig1', 8);
Starting matlabpool using the parallel configuration
'jobmanagerconfig1'.
Waiting for parallel job to start...
Performing parallel job cleanup...
Done.
??? Error using ==>
distcomp.interactiveclient.pGetSockets>iThrowIfBadParallelJobStatus
at 98
The interactive parallel job errored with the following message:

The job was cancelled because the maximum number of MATLAB
Workers available to
run this parallel job, 6,
was smaller than the value of the MinimumNumberOfWorkers
property, 8.
---------------

It is strange because if I run "nodehoststatus" I can see
eight workers connected to my jobmanager but for some reason
I have only access to 6. If I eliminate all the workers in
the 32bits machines and I left only the two workers in the
64bits machine, then I do can use the 2 workers in the
machine. I guess the problem comes because of mixing
different SOs.

Is it normal? am I doing anything wrong? is there any solution?

Regards,

From: Edric M Ellis on 7 May 2008 06:37
"Jaime Laviada" <jlaviada(a)tsc.uniovi.es> writes:

> I am doing some preeliminaries test in a minicluster with
> the distributed computing engine toolbox from matlab 2008a.
>
> My minicluster has three dual core 2 with WinXP 32bits and
> one dual core 2 with WinXP 64 bits. I create two workers per
> machine.
>
> The problems arises when I create the pool with all the workers:
>
> -------------------------
> matlabpool('OPEN', 'jobmanagerconfig1', 8);

Hi Jaime,

"matlabpool" uses our parallel infrastructure, and unfortunately that's
restricted to machines with matching word-size, endianness and OS type (this is
due to restrictions in our underlying MPI layer), as per:

http://www.mathworks.com/products/distribtb/requirements.html

(You can still run "distributed" jobs across all those nodes simultaneously)

I realise it's not ideal, but you should be able to run 32-bit MATLAB workers on
your 64-bit cluster machine, and this should allow you to use all 8 with
matlabpool.

Best regards,

Edric.
From: Jaime Laviada on 7 May 2008 07:38
Thank you, I will run 32bits workers in all the machines.

Edric M Ellis <eellis(a)mathworks.com> wrote in message
<ytwhcdapfgf.fsf(a)uk-ellise-lnx.mathworks.co.uk>...
> "Jaime Laviada" <jlaviada(a)tsc.uniovi.es> writes:
>
> > I am doing some preeliminaries test in a minicluster with
> > the distributed computing engine toolbox from matlab 2008a.
> >
> > My minicluster has three dual core 2 with WinXP 32bits and
> > one dual core 2 with WinXP 64 bits. I create two workers per
> > machine.
> >
> > The problems arises when I create the pool with all the
workers:
> >
> > -------------------------
> > matlabpool('OPEN', 'jobmanagerconfig1', 8);
>
> Hi Jaime,
>
> "matlabpool" uses our parallel infrastructure, and
unfortunately that's
> restricted to machines with matching word-size, endianness
and OS type (this is
> due to restrictions in our underlying MPI layer), as per:
>
> http://www.mathworks.com/products/distribtb/requirements.html
>
> (You can still run "distributed" jobs across all those
nodes simultaneously)
>
> I realise it's not ideal, but you should be able to run
32-bit MATLAB workers on
> your 64-bit cluster machine, and this should allow you to
use all 8 with
> matlabpool.
>
> Best regards,
>
> Edric.

 | 
Pages: 1
Prev: find user and org info for current install?
Next: R2008a Windows Installation Woes