open TELEMAC-MASCARET The mathematically superior suite of solvers

Good afternoon,
I am seeking for a proper way to install and run Telemac 2D on a linux cluster with the following characteristics:
- architecture: 1 head node plus 32 compute nodes (no direct user login on compute nodes permitted)
- scheduler: SunGridEngine (in total 512 slots)
- available compilers: pgi, v. 10.0; gnu, v. 4.3.4
- available MPI implementations: openmpi, mpich2, mvapich2.

We already install telemac2d using the makeall90 & makepar90 files but there are still some problems :
1. calculations sometimes run for ages with providing and results (simulation stucked on the time step 0).
2. the fortran compilation does not work, with the following error message :
"*** EDITION DE LIENS ***

ERREUR A L'EDITION DE LIEN "

In attachment of this post, the job file used to run a simulation via qsub command and the config file of telemac2d.

I hope my post is clear enough.
Thank in adance for your response,
Regards
Renaud

Hello,

Well, hard to say, I give you some details on how we are using clusters at Edf:

+ Except on BlueGeneP - where Perl is not supported on nodes - we always managed to run telemac without breaking the perl sequence.

+ As you did it, mpi_telemac.conf is parsed using nodes adresses given by the scheduler (PBS_NODEFILE or hostfile)

+ OpenMpi has a known "bug" : we weren't able to pass environment variables through QSUB. We were forced to source directly TELEMAC environment (PATH=...bin:$PATH + SYSTELCFG) and OpenMPI module (PATH + LD_LIBRARY_PATH) from our bash_profile/bashrc when computing on node. Error was "orted not found".

+ With OpenMPI, I used RUN_MPI="orterun -hostfile mpirun.txt -np ..." where mpirun.txt is parsed as above. We fill "LK_MPI" with results from "mpif90 -showme:link".

And now some questions :

=> LK_MPI is disabled in your systel.ini. Any reason? If disabled, you must have been in trouble with 'makepar90' (e.g. 'maktel parallel' in the sources)

=> Please give us the complete outputs when compiling the user fortran file. It happened on some cluster that this file was not found. This was due to a perl error, we temporarily replaced some "§" by ";" in $fn=join "§",@evals2[0],@items[1]

" 1. calculations sometimes run for ages with providing and results (simulation stucked on the time step 0) "

=> With both OpenMPI & MPICH library?

=> What about serial computing? Does it work correctly?


I hope it will somehow help.

Best regards,

Fabien Decung