[Wrf-users] DMP vs SMP and ARW vs NMM

[Cory Davis]

Hi there,
I have been trying to assess the performance of both WRF cores on my
dual cpu clovertown (quad core) with various compilation options.  IN
particular I have run ARW with both the SMP(OpenMP+RSL) and
DMP(MPICH+RSL) options appropriate for the intel compiler on a two-way
nested domain.  

There is a huge difference between the multi-process gain between the
two options. mpirun -np 8 is over 2 times faster than the SMP version
with OPT_NUM_THREADS=8.  Inspection with top shows that the MPI option
is almost fully ulitising all 8 cores, whereas the SMP option is using
less than half.

I was wondering whether this is to be expected, whether others have
the same experience, or whether there is some problem with my SMP
compilation/configuration.

Also, the NMM core (MPI, RSL_LITE) seems about 3 times faster than ARW
for both nested and unnested runs.  Is this to be expected?

In both cases I have used the default physics options.

Cheers,
Cory.