[gmx-developers] mpi and thread command line options
David van der Spoel
spoel at xray.bmc.uu.se
Wed Jul 11 09:41:53 CEST 2012
On 2012-07-11 09:12, Berk Hess wrote:
> Hi all,
> Thanks for all the feedback.
> What I forgot to say is that most users will not have to look and
> probably will not look
> at all these options, as we will automate things as much as possible.
> Furthermore, with the group cut-off scheme (the only scheme pre-4.6) openmp
> threads are not even supported, so there is not much option. With the
> Verlet scheme
> combining MPI processes/threads with openmp threads is slower, except at
> the scaling limit. Within a single CPU openmp threads might be faster
> than mpi.
> We can handle all these cases, except for optimization at the scaling
> automatically when using all available cores.
> While thinking of this, and if we don't add mpi spawn, I came up with a
> new proposal:
> -nt total number of threads (effectively the same functionality as in 4.5)
> -ntmpi number of MPI threads (advanced option)
> -ntomp number of OpenMP threads (advanced option)
> In this way the user would not have to worry/know about MPI and openmp
> and would not have to set anything, or only -nt when using part of a node;
> mdrun can choose the optimal number of MPI and openmp threads.
> We might move away from openmp in the next release, but if users would
> never use -ntomp, renaming that option later is not problematic.
For my understanding, is there two-level MPI parallellism in 4.6 or not?
In 4.5 thread-mpi and normal MPI were mutually exclusive at compile
time, is this still the case?
From your mail it seems that there will be a new option combining
node-level MPI with OpenMP. So there can be a trade off between number
of MPI processes and OMP cores/MPI process. I guess that if GPUs are
involved they are tightly coupled to the OMP process, right?
Berk, didn't you have a picture of all the possibilities in your
David van der Spoel, Ph.D., Professor of Biology
Dept. of Cell & Molec. Biol., Uppsala University.
Box 596, 75124 Uppsala, Sweden. Phone: +46184714205.
spoel at xray.bmc.uu.se http://folding.bmc.uu.se
More information about the gromacs.org_gmx-developers