open TELEMAC-MASCARET The mathematically superior suite of solvers

Thanks a mil Sebastien, I will amend my file now and hope it will work.

Kind Regards!

Violeta

Hi Sebastien,

You were right. I looked at CMakeCache.txt and all compilers/links indicated to openmpi_gnu/1.6.4, but should be openmpi_intel/1.6.4. I was not sure if could overwrite it manually and then install metis again. I have not seen any notes on this on google so since gfortran is there by default I went back to gfortran, for fifth time now:). It compiled fine and run examples in scalar on cluster fine as well. But it does not run in parallel. It goes to my ongoing error:
sh: gfortran: command not found

It is weird as gfortran is on path. Well if I type "which gfortran" then it shows me the path: /usr/bin/gfortran. But I have not entered it on PATH myself. Some people lookes at this error and they do not think it is really gfortran issue but I do not know what else it could be.

Also for some other CFG models (like Delft) users have to specify the nodeXX they want to run on but I have not seen any options for Telemac.

Thank you.


Kind Regards!

Violet

Hello,

Have a look at some of EDF's configurations, those using hpc_stdin and HPC queues. You can add the name of your host also. In summary everything is done in the config file.

Sometimes, you need node names when you run on a queue -- by default it uses the local name -- I think there is a bit of a mix in your config.

I note that you have options hpc, but no hpc_hpc_cmdexec in ubuifortopenmpi. And you may wnat to use ubugfopenmpi, but in the hpc_stdin, you have no call to mpiexec -- also I think it is better to use mpiexec than mpirun ...

Finally, we found that we need to change the access right to the submission file with:
hpc_cmdexec: chmod 755 <hpc_stdin>; qsub < <hpc_stdin>

Hope this helps.
Sébastien.

HI Sebastien,

Thanks for your advice. I updated hpc options and amended hpc_cmdexec line as well. I do not run with ifort now as I could not compile so I have not updated hpc option for that. At the moment I use ubugfopenmpi only, that is why I have hpc there. After latest amendments I have not gotten any errors, what is promissing I hope, but I think that my working directory is wrong as it tries to run from my HOME directory. Is there some way I can set up working directory in config file or it is some extra directory to be set up manually? Thanks.

Kind Regards!

Violeta

HI Sebastien,

I have two news, bad and good. The good news is that it is working with hpc options for cases without fortran files, but bad news is that with fortran files it gives me error - sh: gfortran: command not found. So at least I know that my cofnig file is right, I hope. Any ideas what else I can do? Maybe there is some missing directories? I have added /usr/bin/gfortran to PATH already. After so many weeks of trying I made a little step forward:), but not there yet. I attached my latest cofnig file, maybe there are some little errors related to gfortran.

Thanks.

Kind Regards!

Violeta

Hello,

I also have problem when I try to run runcode.py within the queue. I usually do --split and -x on the head server (using the same -w ./workdir option.
then I run it with --run but my stdin only includes the <mpiexec> command. And when it is done I use --merge.

Instead you have this:
/home/violeta.moloney/opentelemac/v6p3/scripts/python27/runcode.py <codename> <casfiles> <options>

I'd advise you use "load module ..." within the queue too, before you launch runcode.py. load module within the queue will ensure the gfortran, mpi, python are fully loaded.


I will talk to my IT next week to investigate that too.

Sébastien.

Thanks a mil Sebastien, I will try do that. I will let you know.

Thanks again and have a nice week-end!

Kind Regards!

Violeta

Hi Sebastien,

Thanks again. I managed to include module load into hpc options, but I am little bit confused in terms of "split" and "-x" and where exactly to add <mpiexec> in hpc options. You mentioned to add mpiexec in hpc options already but I was not sure how and where. Sorry for my lack of programming skills but I want to try it too as it might work for me as well. Thanks.

Kind Regards!

Violeta

Here is a configuration that works for me (note that you are using a different queue system so you have to adapt to yours):


With this you have can either do telemac2d.py -s -w ./trace casfile, and at the end you will have to do telemac2d.py -s -w ./trace casfile --merge

You can also do the first step in several sub-steps:
telemac2d.py -s -w ./trace casfile --split
telemac2d.py -s -w ./trace casfile -x
telemac2d.py -s -w ./trace casfile --run
as long as you keep using the same tmp dir "./trace".

Have a nice weekend.
Sébastien.

Hi Sebastien,

Thanks again for helping me with parallel installation last week. I followed your advice as per your last note and it worked! The second option with sub-steps. So the cases with fortran files worked this way. However, since I run it on 1 node and 1 processor the jobs do not go on gueue but run like in scalar. If I increase the number of processors in cas file then jobs do not run. Funny that when I run jobs with hpc option in command line but without the fortran files then they go on queue. I feel like I am getting closer but for the moment I have:
1) jobs run on queue without fortran files
2) cases with fortran files run with --split, -x, --run and --merge options, but do not go on queue.

I still get the error message 'sh: gfortran: command not found' when I try to run cases with fortran files with hpc option in command line.

Maybe I have to set up some directories there?

Is there anything else I can try, what I have not tried yet :)?

Thanks.

Kind Regards!

Violet