Distributed memory parallelization algorithm (ParMmg)


Mesh adaptation on distributed memory architectures


  1. Mesh distribution: centralized input meshes are partitionned and distribued among available MPI ranks, distributed input meshes are rebalanced;
  2. Mesh subdivision: on each processor, the mesh can be subdivided into sub-meshes (groups);
  3. Mesh adaptation: the sequential remesher Mmg is called on each sub-mesh. Interfaces between the sub-meshes are not authorized to be modified (we say that those interfaces are constrained or frozen);
  4. Interface migration: In order to be able to remesh the constrained interfaces, we create new sub-meshes using a front migration algorithm or a new call to a graph partitionner: interfaces that were frozen during the previous step should be inside the newly defined sub-meshes;
  5. Go to step 3 until reaching the stop criterion (maximal number of iteration).

Comments are closed.