What is the memory overhead introduced for coupling solvers using preCICE?
This consists of many parts:
- The configuration state
preCICE is a configuration-driven library. This involves bookkeeping at run-time. - The mesh and its data
vertices and connectivity information will result in additional memory consumption.
Note that each vertex holds data that needs to be received, mapped, and send. - The mapping methods
Mapping methods strongly vary in memory overhead.nearest-neighbour
andnearest-projection
create a spacial index tree. RBF mappings preallocate additional memory for solving systems of equations.
We are working on a (close-to) zero-overhead solution.