Running parallel
From the start PumMa has been designed as a parallel molecular dynamics code. As of today the parallel version of PumMa scales very well on many processors, making it a very useful program to be used in large scale biological related simulations, such as the transition of large bilayer into a vesicle or the fission of a large vesicle into two smaller ones.
The parallel version of PumMa is based on the Message Passing Interface (MPI). To be run in parallel MPI has to be installed on the system. Depending on the type of installation PumMa can also be run using Local Area Multicomputer (LAM) on top of MPI.
Many small Linux-based beowulf clusters are using the Portable Bash System (PBS) as the queue system to start simulation jobs. Most of our simulations on any of or clusters have been started through the PBS queue system. For these type of clusters it is important to know the amount of memory that is accessible by MPI for sending and receiving information about the particles of the simulation. The P4_GLOBMEMSIZE variable can be used to specify the amount of memory that should be used by the program for communication. This memory size and the number of particles determine to a certain extend the maximum value of the input option Dumpsize. For most systems the default value 100000 of Dumpsize is satisfying.
Since it very much depends on the way the parallel cluster is installed, it is quite hard to give an example how to start PumMa in parallel. However, it is common to distinguish the sequential from the parallel version, by removing the capitals in the program name. Hence the parallel version is most often called pumma.
<< Running sequential | Manual | On screen information >>