Center for Petascale Computing
FLASH's core technology is adaptive mesh refinement (AMR),
which allows high-resolution meshes to be placed in regions of interest. FLASH's AMR capability
is provided by the NASA-supported PARAMESH library, which uses an oct-tree AMR algorithm
that assigns blocks to processors using a space-filling curve, ensuring proper load balance. We
are interacting with CS research by migrating FLASH to the CHARM++ AMPI framework. Using
AMPI gives us an additional load-balancing degree of freedom beyond PARAMESH's space-filling curve technique to address multigrid scaling issues. AMPI also gives us access to the fault-tolerance capabilities of CHARM++ and the Projection performance analysis tool. A Photran based refactoring tool has been
developed to automatically convert FLASH code to use AMPI.
Preliminary results in paper (listed below) demonstrated the usefulness of
AMPI features.
Investigator: Paul Ricker
Astrophysics - FLASH Information