Using DDSCAT with MPI for nearfield calculations: fixed in DDSCAT 7.3.2

Summary:

Using DDSCAT with MPI for nearfield calculations, and release of DDSCAT 7.3.2

Using DDSCAT with MPI for nearfield calculations: fixed in DDSCAT 7.3.2

draine 25 Aug 2016 18:59

Sergey Ph. Ruzankin reported that DDSCAT 7.3.1 failed to properly carry out nearfield calculations (NRFLD=1 or 2) when used with MPI. Dr. Ruzankin also recommended some changes to DDSCAT.f90 to correct this problem.

The changes recommended by Dr. Ruzankin have now been implemented in a new release, DDSCAT 7.3.2, released today (2016 August 25) and available for download from ddscat.org — go to the "Downloads" tab. At this point in time we consider DDSCAT 7.3.2 to be fully compatible with MPI. If users encounter bugs with use under MPI, they should report them here.

Javier (guest) 17 May 2017 09:33

Hello!

I am executing ddscat as a part of a University Thesis and I am having trouble using MPI for parallel calculations. I have followed all the instructions in the User Guide, having changed all the parameters necessary in both files "ddscat.par" and "Makefile". Once compiled, I am having trouble making the code understand it has to separate itself in different cores: it keeps on calculating everything in a single node. How do I manage to tell the code to run separately? Is there a parameter I should change that I don't know about?

Thank you!

draine 17 May 2017 11:49

The first step would be to verify that MPI is installed and operational, and what the local "job submission" procedures are on your system (which will include telling MPI how many cores or cpus to use for the calculation).

Have you succeeded running any other code using MPI on your system?

Javier (guest) 31 May 2017 11:21

Hello,

other codes using MPI have been successfully implemented in our system, therefore ours should be able to work too. Also, the job submission procedures are correctly implemented in our system. We have followed all the instructions in the USer Guide: we changed the PBCGS2 parameter to PBCGST in the file ddscat.par; also, in the "Makefile" we commented the initial parameters and commented the necessary ones:

DMPI = -Dmpi
MPI.f = $ (MPI_f)
MPI.o = $ (MPI_o)

Nevertheless, it still does not seem to work. Should we be changing any other parameters? If not, is there any parameter we stated incorrectly?

Thank you very much again.

draine 01 Jun 2017 15:52

1. The lines mentioned in the Makefile should read
MPI.f = $(MPI_f)
MPI.o - $(MPI_o)

i.e., there should be no space between the $ and (

2. When you compile and link, which of the following do you see used?
mpi_subs.f90 etc. on compilation and mpi_subs.o on linking,
or mpi_fake.f90 and mpi_fake.o ?