With queueing systems

Let's start with the simple and common case of a computer cluster with a queueing system. We don't give here the specific syntax of the job submission script, because it is highly system-dependent. In a nutshell, one would submit on job containing the following commands, assuming the system is using a standard implementation of MPI:

 [some header specifying the number of nodes etc.]
 maps -d & (note the & to run in the background.)
 pollmach runstruct_vasp mpirun (you may need to specify the number of processors as mpirun -np [number])
If your system does not allow a background commands in the script (or if this would be very inefficient), then you could put each command (without &) in a separate script, but making sure they run at the same time. Either way, parallelization occurs at the level of the vasp command. If you wish to also parallelize by running multiple vasp commands simultaneously then you should submit yet another job script with
 [some header specifying the number of nodes etc.]
 pollmach runstruct_vasp mpirun
repeating this process as many times as you the number of simultaneous vasp processes you want (note that there is still only one copy of maps running).

[email protected] Wed, Dec 6, 2023 12:55:16 PM