Dear community,
I am actually not 100% sure this is a problem related to MFIX itself.
I have noticed that parallel simulations on our HPC cluster are slower when using multiple nodes on the cluster. For example, I have run the same MFIX simulation (with the latest version) on the cluster twice: once using 16 cores (which is the number of cores available in each node, hence using one node) and the other using 96 cores, hence 6 nodes.
The first one is always significantly faster than the second one (169 vs. 379 minutes).
At first I thought it might be a problem related to our cluster, but then I run a similar test with another model using mpirun like for MFIX. And for this model the simulation with 96 cores was significantly faster than that with 16 cores. Therefore I am now wondering it may be related to the parallel version of MFIX?
I have used the same compilers for both models (gfortran and openmpi).
Thanks
Fabio