Welcome to the new DiscussCESM forum!
We are still working on the website migration, so you may experience downtime during this process.

Existing users, please reset your password before logging in here: https://xenforo.cgd.ucar.edu/cesm/index.php?lost-password/

How to use mpirun to reduce running time?

majun

Member
Dear scientist,
Now I can successfully run clm5.0 at regional scale and 0.05° spatial resolution. However, it costs lots of time to run. I want it faster but don't konw how to set.
The situtaion of my server is as follows:

Logical CPU Number: 24
Physical CPU Number: 2
CPU Core Number: 6
Memory:128G

I have installed mpich 3.3.1 and make sure it can work . I have installed pnetcdf but don't know how to use it . I didn't use batch system. Meanwhile, I have changed some configuration in config_machines.xml as follows:
....
<MAX_TASKS_PER_NODE>6</MAX_TASKS_PER_NODE>
<MAX_MPITASKS_PER_NODE>6</MAX_MPITASKS_PER_NODE>
<mpirun mpilib="default">
<executable>mpirun</executable>
<arguments>
<arg name="ntasks"> -np {{ total_tasks }} </arg>
</arguments>
</mpirun>
and changed some configuration in config_compilers.xml as follows:
<MPI_LIB_NAME>mpich</MPI_LIB_NAME>
<MPI_PATH>/home/shenhf/majun/setup/mpich-3.3.1</MPI_PATH>
<NETCDF_PATH>/home/shenhf/majun/setup/netcdf</NETCDF_PATH>
<PNETCDF_PATH>/home/shenhf/majun/setup/pnetcdf-1.12.0</PNETCDF_PATH>

However, when I execute ./case.submit ,it always shows:
mpirun -np 1 /home/shenhf/majun/clm5.0/output/80x60_SP_spinup/bld/cesm.exe >> cesm.log.$LID 2>&1
when I use mpirun -np 6 /home/shenhf/majun/clm5.0/output/80x60_SP_spinup/bld/cesm.exe it shows:
ERROR: (cime_cpl_init) :: namelist read returns an end of file or end of record condition
#0 0xa3a7f9 in ???
#1 0xa3a9c6 in ???
#2 0x428af4 in ???
#3 0x42b194 in ???
#4 0x31fe61ed5c in ???
#5 0x4104e8 in ???
application called MPI_Abort(MPI_COMM_WORLD, 1001) - process 0

So how to set to enhance the efficiency of run? Any suggestion or help is highly appreciated.
 

sacks

Bill Sacks
CSEG and Liaisons
Staff member
I think what you're asking is how you can use more processors for the run. After setting up the case, you can run:

Code:
./xmlchange NTASKS=6

so that the run uses 6 processors.
 

majun

Member
I think what you're asking is how you can use more processors for the run. After setting up the case, you can run:

Code:
./xmlchange NTASKS=6

so that the run uses 6 processors.
thanks a lot. I have another question on how to create irregular surface datasets using a mask(like shp file). Now I can create regular(rectangle) surface datasets using some scripts attacted with clm5.0(e.g.mknoocnmap.pl, mkmapdata.sh gen_domain etc.), however my study region is irregular, what should I do to ? Is there any instructions to follow? Any suggestion or help is highly appreciated.
 

sacks

Bill Sacks
CSEG and Liaisons
Staff member
Please open a new thread for your new question, posting it in the land forum. When you do, please describe your question in more detail.
 
Top