: derived ParallelDirectApplicInterface class declaration and implementation, taylored to ISSM. This class is registered into the interface database of Dakota, and is used to directly call ISSM cores from Dakota. More...

Detailed Description

: derived ParallelDirectApplicInterface class declaration and implementation, taylored to ISSM. This class is registered into the interface database of Dakota, and is used to directly call ISSM cores from Dakota.

. This code is only valid for Dakota versions higher than 6!

This routine helps running ISSM and Dakota in library mode, for Dakota versions that are >=6, and that fully support parallelism. The setup is radically different than from version <6! Now, dakota runs the show more. The reason is that dakota now controls the parallelism in a master/slave setup, and hands over to ISSM a bunch of slave communicators, which are then used to run our simulations. Because ISSM is now ESMF compliant, we can use these communicators to create separate identical FemModel instances on each slave communicator! This allows us to scale to large jobs (think 1000's of cpus), which we split into multiple sub-slave communicators, which run the sampling (or forward different, local reliability, optimization you name it) simulations on each slave.

This is all bootstraped from the main issm_dakota main, (see c/main directory), which is heavily inspired on the main found in the dakota/src/library_mode.cpp code. We also have to create an ISSM code that registers into the dakota database, which is capable of running ISSM. This is derived from the Dakota class called ParallelDirectApplicInterface.

Definition in file IssmParallelDirectApplicInterface.h.