||October 29, 2009 06:17
Load balancing pre- and during computation
I'm currently using decomposePar to decompose the mesh into n portions for running in parallel. The actual mesh partitioning itself appears very efficient and there is almost an equal number of cells on every node. However, when I'm actually running the solver (in this case, it is just potentialFoam), it appears to overload the memory distribution on the first processor, then the second, then the third, and fourth, each time removing memory allocated to other processors... is there a reason for this, or have I missed out a step?