question about MPI
Lo-hatch opened this issue · 1 comments
I have two question about MPI:
-
If I run ED on a single plot, can I use more than one node to accerlate simulation?
when I use 32 cpus in one node, the time sequence in log file is normal, however, when I use two nodes, it become choatic.
Most confusing thing is the result of the two simulation is totally the same. -
If I run ED on a single plot, I don't need and should not use "mpirun -np 32", this is only used for multiple-grids run?
MPI is only used for mutliple gridcell (or polygon in ed2 terminology) run. Each grid cell will be assigned a single core.
For single grid/polygon/site simulation, you can use openmp, in this case, the model will distribute patches to different number of cohorts. The maximum openmp threads can be set through environmental variables.
openmp and MPI cannot be used together at least for now.
It would be great if you can upload the .out file before I can better answer your question