Prospective research areas of LAMMPS

Dear LAMMPS users and developers,

I am a parallel programming researcher and would like to contribute to LAMMPS molecular dynamics software by helping to nail down any bottlenecks that occur in scaling of the software on multiple CPUs (and probably improved performance in exa-scale era.) I am already in process of collecting profiling results to identify the phases that can be improved.

Hence, I would request all of you to please suggest me some possible areas of research, on which we can work on, for better scaling of LAMMPS, (and/or MD softwares in general.) You could also direct me to the link on the website, or any person concerned with this. Any more insight into matter will be really appreciated.

Thanks in advance.

Stan can tell you about Kokkos-related efforts with LAMMPS for

future hardware. The benchmarks page of the web site has

some details.

Steve

Dear LAMMPS users and developers,

I am a parallel programming researcher and would like to contribute to
LAMMPS molecular dynamics software by helping to nail down any bottlenecks
that occur in scaling of the software on multiple CPUs (and probably
improved performance in exa-scale era.) I am already in process of
collecting profiling results to identify the phases that can be improved.

Hence, I would request all of you to please suggest me some possible areas
of research, on which we can work on, for better scaling of LAMMPS, (and/or
MD softwares in general.) You could also direct me to the link on the
website, or any person concerned with this. Any more insight into matter
will be really appreciated.

this is not so easy​ a request to answer. there are multiple ways to
improve performance of any MD code, but there are also other considerations:
- performance optimizations may be very specific for a given model or
system setup
- performance optimizations may interfere with the flexibility,
extensibility and maintainability of the code
- the basic MPI based parallelization strategy using domain decomposition
is sound, and there is very little that can be done to improve this in a
generic fashion, especially for conventional condensed phase systems.

possible routes for improving LAMMPS' parallel performance are in:
- improving the KOKKOS module that steve already mentioned
- improving the load balancing with timer feedback and extending the
recursive bisectioning to more cases
- improving internal data structures for faster data access and better
multi-threading performance
- adding multi-threading and optimizations to "complex" models that don't
support this yet.

HTH,
    axel.

I would like to get in touch with Mr. Stan about Kokkos-related efforts.

Thanks again for the reply!

I will look into all the possible routes suggested. I would like my research to be more generic, rather than being specific (to certain given models or systems), so as to benefit the LAMMPS and MD community in general.

Thanks again for the reply!