I was having some problems with the equilibration of bilayer models from CHARMM-GUI using the scripts provided via parallel CHARMM runs, and discovered the problem was memory usage. It took a while to discover the cause, as the error messages were cryptic, such as:
APPLICATION TERMINATED WITH THE EXIT STRING: Killed (signal 9)
or this one:
= BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
= PID 18876 RUNNING AT i15
= EXIT CODE: 9
The simplest solution is to use fewer cores per node; the number will depend on the node memory per core. Our cluster has most nodes with 2 GB/core (e.g. 16 cores with 32 GB memory), and the processes were using well over 3 GB each, so using 8 cores per node solved the problem.
If the molecular system is modest in size, one could also disable (via ! char) the following line at the beginning of each script:
DIMENS CHSIZE 3000000 MAXRES 3000000
This would reduce memory usage and allow the use of more cores, speeding up the calculation.