Help using GRADS in parallel

[Arlindo da Silva]

Pierre,
    We need to know the system you are running on. Version 1.8SL11 is old
(ca. 2002?) and unless you redid your own build, chances are it is getting
"out of sync" with the underlying system libraries. Upgrade if you can.
GrADS v2 is getting more and more stable with each new release. The builds
found at http://opengrads.org includes some useful extensions (e.g.,
regridding, shapefile support, poisson solver, spectral filters, etc) not
yet found at the official COLA releases.

   About parallelization, again you did not give us much detail. How exactly
did you start your 2 copies of GrADS? I m assuming you are on the same
*node*, attempting to start GrADS on different *cores*. In this case you may
be sharing the system memory with each instance of GrADS you run. Any chance
that you could be running out of memory? There are known memory leaks in
v1.8SL11, specially with templated netcdf/hdf-4 files.

  About starting several instances of grads concurrently. There are several
options to do that. Depending on your OS, starting 2 versions of grads in
the background from the shell (I am assuming a Unix like OS here)

% grads -lc 'run script1.gs' &
% grads -lc 'run script2.gs' &

will start each copy on separate cores. SGI's AIX on Altix processors would
even dish out processes across nodes this way; most commonly the
parallelization is done only within a single node. For going beyond a single
node I've written a simple perl client/server script to go across nodes
using mpirun (grads itself does not make use of MPI). I am attaching the
perl script for reference; type prund.pl without arguments for a manual
page.

If you are a python programmer, you could also also use the python thread
class together with pygrads for starting concurrent grads instances.

   Good Luck,

       Arlindo


On Fri, Jul 10, 2009 at 2:36 PM, <pierre.lacombe at mddep.gouv.qc.ca> wrote:

>
>
> I work for the Centre d'Expertise Hydrique du Québec (CEHQ,
> http://www.cehq.gouv.qc.ca), which is an agency of the Québec's ministry
> of environment (formely the Ministère du développement durable et des parcs
> du gouvernement du Québec). Among various objectives, the CEHQ operates the
> hydrometric network of the province of Québec and supports the management
> of public dams by producing hydrological forecasts.
>
>
>
> Over nearly ten years, our hydrological forecasts have been produced dayly
> at a 3h timestep by integrating meteorological forecasts data of the
> canadian model GEM.  To process the GEM input data, we use the GRADS
> system  (Grid Analysis and Display System (GrADS), version 1.8SL11. We run
> GRADS on a serial computer, an it works perfectly well. However, recently we
> aquired a few quadcores computers and found a way to make GRADS work in
> parallel for us. The idea is simple, and consists to let GRADS 1 computes
> the temperature mean, while GRADS 2 computes the precipitation in the same
> time. As a result, we save half of the computing time every morning. This
> work almost fine, except that in certains executions we have an error of
> initialisation of GRADS or GRADS sometimes freezes. So we are unable to use
> this idea for now.
>
>
>
> We wonder if there is a version of GRADS that works in parallel, or are
> there some limits to the use of GRADS that we should know, and that we could
> overpass with some adjustments.
>
>
>
> We thank you in advance for any technical assistance you may kindly provide
> to us, or for your help by transmitting this message to a colleague that may
> better answer to these questions.
>
>
>
> Best Regards,
>
>
>
> Pierre Lacombe, computer analyst
>



-- 
Arlindo da Silva
dasilva at alum.mit.edu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://gradsusr.org/pipermail/gradsusr/attachments/20090711/a6acf49d/attachment.html 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: prund.pl
Type: application/octet-stream
Size: 5150 bytes
Desc: not available
Url : http://gradsusr.org/pipermail/gradsusr/attachments/20090711/a6acf49d/attachment.obj