problem of parallel job with TURBOMOLE5.10 on x86_64 machine

[Xiaoyan]

Hi, I am trying to run parallel job on x86_64 machine, then I get the below error message:

/scratch/cao/TURBOMOLE/bin/x86_64-unknown-linux-gnu_mpi/ridft_mpi: symbol lookup error: /scratch/cao/TURBOMOLE/bin/x86_64-unknown-linux-gnu_mpi/ridft_mpi: undefined symbol: __pgi_abort_init
MPI Application rank 2 exited before MPI_Init() with status 127

[uwe]

Hello,

the binaries are dynamically linked. If you start a parallel application, the binaries as well as the shared libraries in $TURBODIR/libso have to be accessible and readable on each node. Did you install Turbomole on an NFS disk, i.e. is /scratch/cao/TURBOMOLE in your case a directory that is mounted on all nodes?

Uwe

[Xiaoyan]

Dear Uwe
Thank you for your prompt reply--- the Turbomole5.10  is installed on the local disk of the computer where I run the job, I have checked the  directory /scratch/cao/TURBOMOLE/libso by using the "ls -l", then I get the below message:
drwxr-xr-x  5 cao users 4096 2008-01-09 15:52 libso
the subdirectory x86_64-unknown-linux-gnu_mpi has the same properties as libso. do you see anything wrong ?

Actually the TURBOMOLE have been installed on an NFS disk too, but I got the same error message if I started my job by using the installed TURBOMOLE on NFS.


Kind regards
Xiaoyan

Hello,

the binaries are dynamically linked. If you start a parallel application, the binaries as well as the shared libraries in $TURBODIR/libso have to be accessible and readable on each node. Did you install Turbomole on an NFS disk, i.e. is /scratch/cao/TURBOMOLE in your case a directory that is mounted on all nodes?

Uwe

[uwe]

Hi,

the library that is not being found should be:

/scratch/cao/TURBOMOLE/libso/x86_64-unknown-linux-gnu_mpi/libpgc.so

Try to set

export LD_LIBRARY_PATH=/scratch/cao/TURBOMOLE/libso/x86_64-unknown-linux-gnu_mpi/

and rerun the job. All libraries needed for calculations are included in the directory given above.

Regards,

Uwe