[gmx-developers] CVS implementation of REMD

[Berk Hess]

Ken Rotondi wrote:

> Hello all,
>
> I'm having a bear of a time getting started with REMD in the CVS code.
>
> I have succeeded in compiling the CVS code with mpi support.
>
> I've made two .tpr files for two adjacent temperatures, named 
> topol0.tpr and topol1.tpr  as an initial test, using:
>
> grompp -v -np 2 -f 1.mdp -po 1out.mdp -c md8.gro -p p39a.top -o 
> topol0.tpr
>
> To start mdrun:
>
> mdrun -v -multi -np 2 -replex 20 -s topol*.tpr -o
>
> Fatal error: run input file topol0.tpr was made for 2 nodes, while 
> mdrun expected it to be for 1 nodes
>
> mpirun -np2 mdrun -multi -v -np 2 -replex 20 -s topol*.tpr -o
>
> returns the mpirun man page
>
> This is a single processor box, but I don't see that effecting the 
> number of processes that can be run (just the speed with which they 
> will finish).
>
> What am I doing wrong?
>
> How are the input and output files specified in the mdrun command 
> (using a * wildcard)?
>
> Unfortunately the CVS manual does not include the Appendix with the 
> manual pages in it, however, mdrun -h should provide the same 
> information.
>
mdrun -h gives the required help.
The -replex help refers to -multi, which in turn says:

With -multi multiple systems are simulated in parallel. As many (single 
node)
input files are required as the number of nodes. The node number is appended
to each input and output filename, for instance topol.tpr becomes 
topol0.tpr,
topol1.tpr etc. The main use of this option is for NMR refinement: when
distance or orientation restraints are present these can be ensemble 
averaged
over all the systems.

So you did two things wrong.
The tpr files should be single processor (grompp -np 1).
And you do not need to specify -s (or specify -s topol.tpr).
Maybe we should clarify the second point in the help.

Berk.