Hello, after checking the FEATool website, I can see that it is possible to use parallelisation for the actual computing, but is it at all possible to do distributed computing ? IE using several computers linked up to a main node instead of one computer ?
|
Administrator
|
Thank you for getting in touch and interest in the toolboxes. Regarding distributed simulations, the toolboxes support external solvers OpenFOAM, SU2, and FEniCS which all support distributed solution processes via MPI parallelization. However, launching distributed simulations, control, and processing has not been implemented yet. For now only solving in parallel is only supported on one shared memory system with the built-in functionality (MPI + OpenMP). So, technically the toolboxes can be used to define, mesh, and setup simulations for distributed computing, but the actual launching/solving has to be performed externally. As the toolbox is fully scriptable with Matlab, this can possibly be done with some custom script functions, but it depends a little how your distributed cluster is set up and works. |
what means "cluster is working properly". There aren't many options there. I have now created a cluster using OpenMPI on ArchLinux. Share SSH keys with each other so that ssh is possible without a password and mount a share folder using NFS and store it in the fstab on the clients (explained to the minimum, write me a private notice if you have any questions). What exactly do you mean by custom script functions and "has been performed externally", can give here an example? Matlab is not well suited for parallel computing. Of course, if you have infinite licenses, but I, for example, only have two licenses. I'm in the process of getting GMSH up and running via OpenMPI. best regards CFD2D
Marten Mehring
Madrigal Industrial Solutions UG |
Does anyone know how to use gmsh here using OpenMPI? I set up OpenMPI and everything works. GMSH also runs with more threads using -nt (number of threads), but how do I use all my nodes with OpenMPI so that I can use even more threads. I've already written a hostfile that addresses all threads via mpirun -np (number of threads). Best Regards CFD2D
Marten Mehring
Madrigal Industrial Solutions UG |
Administrator
|
Haven't really experimented with running Gmsh in parallell. It might be better to ask in the Gmsh forum https://gitlab.onelab.info/gmsh/gmsh/-/issues
|
Free forum by Nabble | Edit this page |