Problems running MPI version of MRCC after MOLPRO

Dear Valérie,
The problem is that only the mrcc executable should be run with mpirun, that is, the last line of your script should read as
mpirun -np $NCPUS mrcc

I changed the script line to
mpirun -np $NCPUS mrcc, but the goldstone error remains!

Can you help me solving this error?

Valérie

Dear Valérie,
Unfortunately I cannot reproduce the problem, the job is running smoothly on our machines. It is probably a compiler-related issue. What compiler do you use?
I would suggest using the precompiled goldstone binary that you can download from this homepage. Just overwrite your goldstone with this binary, and try to run the job. (Please do not change your mrcc executable since it must be compiled for MPI execution, and the mrcc binary provided here is not compiled for MPI.)

Dear Mihály,

I have now used the goldstone binary from your website. It runs fine.
The MPI execution of mrcc faces an error.
Starting CC iteration...
======================================================================
forrtl: No such file or directory
forrtl: severe (28): CLOSE error, unit 21, file "Unknown"

I am using your latest tarball, and I compiled it with the mpiifort compiler version 14.0.4.174
mpiifort -V
Intel(R) Fortran Intel(R) 64 Compiler XE for applications running on Intel(R) 64, Version 14.0.3.174 Build 20140422
Copyright (C) 1985-2014 Intel Corporation. All rights reserved.

Can you please check that error?
Another question, do you have benchmark results for the scaling of MRCC with respect to openMP or MPI parallelization. How many processors can one efficiently use?

Thanks for your help,

Valérie

Dear Valérie,
That is a strange error, and I could not reproduce it, though I tried with another compiler since we do not have that mpiifort version. I am rather puzzled. Probably something is wrong with the read/write permissions on your compute nodes.
Concerning parallelization efficiency: the OpenMP parallelization is quite efficient, especially for perturbative methods. We have tested for up to 48 threads, and the parallelization efficiency was good. With MPI the parallelization efficiency is also fine for perturbative methods, but for the iterative ones it is only satisfactory for very large jobs. Thus for your calculation, which is relatively small, I would not try more than 4 cores with MPI.

Dear Mihály,
Tell me which version of intel compiler you use and produces normal results. I can try it on my cluster.

Valérie