Using multi-node super computer with distributed memory
Hi, I have a super computer with 7 computation nodes each contains 20 CPU processors, hard disk, and 4 GB memory (in total 140 CPU). I know I can't use openMP for this system, I am wondering if there is anyway I can run Dualsphysics parallel on this system.
Comments
Have you managad to run DualSPHysics on mutiple nodes?
I would like to do the same.
The current DualSPHysics code has openMP implemented and not MPI. The use more than a single processor (so far I understand it), MPI is required. When the current code is executed using MPI it is simply executed on each node. This means, if you try to use 20 nodes, the code will be run 20 times, 1 simulation per node. This means no parallelization of the simulation is achieved. So to be able to use multiple CPU's MPI has to be implemented (which is not that easy).