[gmx-developers] IMD patch
martin.hoefling at gmx.de
Thu Aug 28 23:23:08 CEST 2008
Hello Gromacs Developers,
here's my first version of a patch, implementing the IMD = Interactive
Molecular Dynamics Protocol into recent Gromacs CVS versions.
Currently, the following works, at least here:
-Connecting via VMD to a running single or MPI simulation with md integrator.
-Updating the coordinate/energy to vmd and set update frequency from vmd.
-Detaching and reattaching from/to a running simulation.
-Reading and printing out new forces from VMD (pretty useless atm).
When you've compiled a mdrun binary, you can run it with -imd switch. In VMD,
you need to load the corresponding structure and then connect to gromacs via
Extensions->Simulation->IMD Connect (NAMD)
ToDo & Questions:
- My current implementation syncronizes IMD (e.g. output frequency) only on
next IMD output, to reduce communication. Is this ok? My idea would be to
implement the force synchronization from master to other processes here as
-Implement action of VMD forces. VMD sends new forces, applied via mouse to
the master. The simplest idea would be to create a local list with all added,
changed and deleted forces and communicate this list to all processes as
mentined above. Then I need to find out how/where to implement the force
update per process.
-PBC treatment, currently molecules look not so nice after crossing the
-Allow only a certain group to be send to VMD, maybe write out a matching
structure for IMD-client as well.
-Implement IMD into other integrators.
-MPI_Bcast crashes If only one MPI process...
... is (cr->nnodes > 1) the proper condition?
-Energies displayed in VMD might be wrong, conversion should be OK but I
didn't really think about if the contributions are correct.
OK, any comments, suggestions,... ;-)
...I am sure there's a lot of stuff one can do better.
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 15073 bytes
Desc: not available
More information about the gromacs.org_gmx-developers