Running LES in multiprocessing mode

This post was from a previous version of the WRF&MPAS-A Support Forum. New replies have been disabled and if you have follow up questions related to this post, then please start a new thread from the forum home page.

H

HP1801

New member
I am trying to run WRF-LES model and have compared running it with a single processor using
Code: 
./wrf.exe
and in multiprocessing with 6 cores using
Code: 
mpirun -np 6 ./wrf.exe
, however, I did not notice any differences in run times for the simulator. Is this normal or am I missing something?
 
Hi,
Just to verify, did you compile the model 2 different ways - serially, and then with dmpar and an MPI-type library (for parallel computing)?

If so, can you do the same test again and when running it serially, send the output to a log file:
Code: 
./wrf.exe >& log.wrf

then please attach that log.wrf file, along with the rsl.out.0000 file from your parallel run, along with your namelist.input file so that I can take a look? Thanks!
 
kwerner said:
Hi,
Just to verify, did you compile the model 2 different ways - serially, and then with dmpar and an MPI-type library (for parallel computing)?

If so, can you do the same test again and when running it serially, send the output to a log file:
Code: 
./wrf.exe >& log.wrf

then please attach that log.wrf file, along with the rsl.out.0000 file from your parallel run, along with your namelist.input file so that I can take a look? Thanks!
Click to expand...

During ./configure, I picked option 35 (dm+sm) for GNU (gfortran/gcc). I am not sure which library from the list is for multiprocessing. From what I understood dm+sm should support both? Sorry I am a bit new to this.
Do you mean I need to compile the library serially first and then with dmpar option separately?
 
Hi,
dm+sm is a distributed memory, plus shared memory option for compiling. We actually don't usually recommend using the dm+sm option because it has caused many people problems, and we don't test it here. I would advise compiling with only dmpar (distributed memory). Then you can run a test with 1 processor, and again with 6 processors to see if you get timing differences. When running with each, use the mpirun command, but just change the number of processors:
Code: 
mpirun -np 1 ./wrf.exe
Code: 
mpirun -np 6 ./wrf.exe

You can compare the times you find in the rsl.out.0000 files for each of these tests to see timing results.
 
kwerner said:
Hi,
dm+sm is a distributed memory, plus shared memory option for compiling. We actually don't usually recommend using the dm+sm option because it has caused many people problems, and we don't test it here. I would advise compiling with only dmpar (distributed memory). Then you can run a test with 1 processor, and again with 6 processors to see if you get timing differences. When running with each, use the mpirun command, but just change the number of processors:
Code: 
mpirun -np 1 ./wrf.exe
Code: 
mpirun -np 6 ./wrf.exe

You can compare the times you find in the rsl.out.0000 files for each of these tests to see timing results.
Click to expand...

Alright I will try this, thank you so much! I I still have problems I will let you know.
 
You must log in or register to reply here.
Back
Top