[DFTB-Plus-User] Regarding DFTB+ in parallel

[Ben Hourahine]

Hello Hegoi,

at the moment DFTB+ is openMP parallel but not MPI parallel
(so don't use mpirun, just start the binary).

To use all of the cores on your system you will need a threaded
blas library. MKL is threaded as is ATLAS (if compiled correctly).
You also will need to set the shell variable OMP_NUM_THREADS
to 16.

Also,  to get any benefit you'll need to run problems which are large
enough that the parallelism gain exceeds the overhead of using parallel
algorithms. This depends on your hardware, but I'd expect system sizes
of several hundred atoms before the linear algebra problems become
large enough to gain from parallelism.

Regards

Ben

Hegoi Manzano wrote:
> Hi,
>
> I have compiled DFTB+ in a Dell x_64 machine with 4 processors each of them with 4 cores. I want to run the job in parallel using the 16 nodes that I have. Then, how should be the mpirun command? If I try
>
>  mpirun -cpus-per-proc 4 ./dftb+
>
> It works in 4 processors, but what about the 4 cores per processor? Are they being used? Sorry for my ignorance on the parallel issues, and thanks in advance.
>
> Hegoi
> ________________________________________________
>
> Dr. Hegoi Manzano
>
> Concrete Sustainability Hub.
> Massachusetts Institute of Technology, Room 5-330
> 77 Massachusetts Avenue, Cambridge, MA02139
>
> e-mail: hmanzano at mit.edu // hegoi.manzano at gmail.com
>            www.hegoi-manzano.blogspot.com
> ________________________________________________
> _______________________________________________
> DFTB-Plus-User mailing list
> DFTB-Plus-User at dftb-plus.info
> http://www.dftb-plus.info/mailman/listinfo/dftb-plus-user
>   


-- 
     Dr. B. Hourahine, SUPA, Department of Physics,
   University of Strathclyde, John Anderson Building,
          107 Rottenrow, Glasgow G4 0NG, UK.
    +44 141 548 2325, benjamin.hourahine at strath.ac.uk
   The University of Strathclyde is a charitable body,
registered in Scotland, with registration number SC015263