Firefly and PC GAMESS-related discussion club

Learn how to ask questions correctly

Alex Granovsky
gran@classic.chem.msu.su

>While Davide is correct that I was more
>wondering about the SMP (multithreaded) aspect of things rather than
>parallel jobs via MPI, the need for the procgroup file can't be stressed enough. (I mostly use openmpi for other programs, and there
>a procgroup or similar file isn't needed when running on a multicore/SMP machine ...)

Well, procgrp file is always needed for MPICH binaries running in parallel;
alternatively, you can install and configure the complete
MPICH-1 installation and use MPICH' mpirun to launch Firefly.
In the latter case, the procgrp file will be created automatically.

>However, what took me longest about getting MPI parallel runs to work, was finding out that the input file (-i input.file) needs to
>be in the first working directory, rather than the directory one
>starts the job from (this applies even more so to the otherwise
>neat -t option).

Indeed, we do not take care of considering/handling the original working directory as the source of input files. This could be
changed in the future; however, all the documentation explicitly suggests to use fully qualified file names.

>Maybe I am just to stupid to read the documentation, but this cost me a lot of time for trial and error, since the error messages are not very helpful. It certainly violates the principle of least surprise: the serial job runs OK, the parallel (MPI) job dies immediately with a very vague error message ...

OK, if you have suggestions on how the documentation could be
improved, or how to improve error messages, please send us
your changes and we'll incorporate them into existing
documentation or code.  

Regards,
Alex Granovsky