Installation of Code_saturne v.2.1.0 on a cluster

[Erwan Liberge]

Hi,
 
i'm installing the 2.1.0 version on a PBS cluster. The installation seems to be right (I have just force the MPI library using LIBS=-lmpi for the compilation ) but it doesn't work.
If i check the mesh using the gui (calculation environnement -> mesh quality criteria), it works, but not for the complete computation (Calculation management-> prepare batch calculation-> run type : Mesh Quality criteria).
I have the following error :
/Calcul/Apps2/saturne4.2.4/v210/code_saturne-2.1.0/libexec/code_saturne/cs_preprocess: error while loading shared libraries: libmpi.so.4: cannot open shared object file: No such file or directory Error running the preprocessor. Check the preprocessor.log file for details.  Error in preprocessing stage. If i cancel MPI, another link is missing.
There is no preprocessor.log file in the result folder (only the xml file, the summary and run_solver.sh).
 
I have another question : how to run a case using PBS cluster ?
 
Regards
 
Erwan

[Yvan Fournier]

Hello,
What OS distribution and MPI library are you using ? How did you configure the code ? Did you use CC=mpicc, --with-mpi=..., or some other means ? ). As you already have an issue with the preprocessor, I assume you used CC=mpicc) With some environments, using "export LD_LIBRARY_PATH=<path_to_mpi_libs>:$LD_LIBRARY_PATH" in your environment may be necessary to avoid this issue.
To use PBS, see the post-install section in the installation documentation (edit etc/code_saturne.cfg or $HOME/.code_saturne.cfg).
Best regards,
  Yvan

[Erwan Liberge]

Thanks for the answer.
 
I have another problem if i add files in the SRC folder. The compilation is done via a sh script, but the compiler (ifort) is unknown (only for a bash script). Using the intel compiler or not is choosed by loading modules, which is not able under sh environment. 
I have tried to add
export ENV=/.../.../.bashrc
in the runcase, but it didn't work.
 
Is there another solution than replace before the installation #! /bin/sh by #!/bin/bash ?
 
Best regards,
 
Erwan

[Yvan Fournier]

Hello,
This seems surprising, as compilation is done through a Python script (which does call a compile command as if it were a shell script), and we use environment modules for the Intel compiler on at least 2 clusters with no issues.
Normally, version 2.1 should keep track of modules loaded at install time, and the run_solver.sh which is generated and used for a run reloads those modules when running (after purging others), but it is possible that recent versions of the modules command are not detected. Version 2.1.1 (released yesterday) should fix this, so you may want to upgrade.
Another possibility is that run_solver.sh uses sh instead of bash, but this would seem surprising (it would happen if the SHELL environment variable were not set).
Best regards,
  Yvan

[Erwan Liberge]

Hello,
The main issue is that module command is unknown in sh environment, but is known in bash ..... but not everywhere ......
 
I will try the version 2.1.1
 
Best regards,
 
Erwan