A priori evaluation of the time / amount of work needed by Mmg to adapt an input mesh and metric

[Algiane]

At each iteration of ParMmg we have to wait until all groups are adapted by Mmg. If work is not well balanced, we end up with MPI processes waiting.

As load-balancing is less important than forcing the interfaces migration (because non adapted interfaces lead to additional iterations), we would like to have the best load balancing that guarantees the migration of interfaces possible.

For now, load-balancing tries to balance the input number of element. We may want to find a better approximation of the workload from the input mesh and metric. We may want to investigate several tracks:

• evaluation of the targeted number of elements
• difference between input and targeted number of elements
• map of area that needs to be refined and area that needs to be coarsen and evaluation of the adaptation time (coarsening takes longer than refinement)
• etc.