IMP logo
IMP Manual  for IMP version 2.14.0
Parallelization

There are several ways to run calculations in parallel in IMP:

  • IMP.mpi module. This is used to run tightly-coupled calculations in parallel using MPI, over processors that are not necessarily on the same machine. In particular, sampling protocols in the IMP.pmi module use this where available. IMP must be compiled with your MPI compiler to use this module. See IMP.mpi for more details.
  • IMP.parallel module. This is designed to manage multiple independent tasks spread over a pool of available workers. The tasks cannot communicate with each other, but the system is fault tolerant and will restart any failed tasks. This module is available in all IMP builds. See IMP.parallel for details.
  • OpenMP support. This is used to spread out IMP's computation between multiple threads on a single multi-core machine. IMP must be compiled with OpenMP support to use this, but when enabled it is used automatically by all parts of IMP that have been parallelized. See the OpenMP page for more details.