Hi Chris,
Welcome to the relax mailing lists! I would guess that the problem
you are seeing is probably more related to the MPI installation and
configuration on your cluster. It may also relate to the mpi4py
site-package you have installed for Python. Could you copy and paste
the output of the relax command:
$ relax --info
This might help in identifying the problem. Have you run the mpi4py
tests (http://mpi4py.scipy.org/docs/usrman/install.html#testing)
to check if the problem is on the cluster, the MPI installation or
mpi4py? When running on my dual core system I could see, in the past,
using 'top' that both cores are being used. I would also guess that
the problem is not in relax as Gary Thompson's multi-processor code
(in the multi/ directory) does not distinguish between cores and nodes
as this is all handled automatically by the MPI protocol. The text
'node' is not even used in this code. But that does not mean that the
problem is not in relax.
Regards,
Edward
On 23 March 2013 16:52, Waudby, Christopher <c.waudby@xxxxxxxxx> wrote:
Hi,
I'm trying to run a model-free calculation in relax 2.2.3 on a cluster, set
up with mpi4py and Open MPI as described in the documentation. I ran the
job with 64 cores, and this was successfully picked up by the software:
Processor fabric: MPI 2.1 running via mpi4py with 63 space processors & 1
master. Using Open MPI 1.4.1.
However, when the job finished an analysis of the CPU time used showed that
only 11 or 12 cores were actually being used for computations, which I
presume correlates with the number of cores per node. Indeed, when I ran
another test using 12 cores over nodes containing 4 cores, only 3 cores
were active during the calculation.
Is there an error in my set up, or is this some kind of bug?
Best wishes,
Chris Waudby
c.waudby@xxxxxxxxx
_______________________________________________
relax (http://www.nmr-relax.com)
This is the relax-users mailing list
relax-users@xxxxxxx
To unsubscribe from this list, get a password
reminder, or change your subscription options,
visit the list information page at
https://mail.gna.org/listinfo/relax-users
Re: relax on cluster