ggi and cyclicGgi performance
I'd like to compare my experience with GGi and cyclic GGi. I usually apply these interfaces to the MRFSimpleFoam solver (a marine propeller case). The simulation runs on a cluster: every node consists of 2 Xeon E5520 processors (4 cores, 2.26 GHz) and 12 GB RAM. As I noticed the solving was not faster with a larger number of nodes, I made some little trials changing the CPUs number. I report my results in the attached graph (every line refers to the number of first iterations). I obtain the best performace with an almost-1 node (7 CPUs): is it common? I have no such experience to understand if it depends on the available hardware, my case settings or a (non-)optimization of ggi and cyclicGgi algorithms. I'd like to know your opinion about that.
|All times are GMT -4. The time now is 19:42.|