Poster by: Louis Moresi, School of Geosciences and School of Mathematical Sciences, Monash University, Clayton, VIC, Australia.

We will share our experience in re-implementing a widely used geodynamics-modelling research code as a community resource for deployment in massively parallel, high-performance computing environments.

The community imposed a number of additional requirements on the new code:
• Parallel efficiency
• Versatility in tackling new problems
• The capacity to integrate multiple models
• Modular development environment with sandboxes for new functionality
• Integration with standard packages for grid deployment

The methods employed in the code require both distributed implicit equation solves and the tracking of unstructured swarms of material points through a decomposed domain. We faced a number of challenges in creating reliable, efficient and comprehensible implementations of these algorithms.

We have constructed our solver technology upon standard parallel solver packages such as PETSc from Argonne National Labs, as componenet which can be switched at run-time. Rheology, mesh geometry, boundary conditions, meshes, particle-swarms are all implemented in the form of exchangeable components to allow users to compose problems freely and to add / exchange their own capabilities.

In a companion poster (The Underworld of the ARCS grid by Mason, Binsteiner, Quenette, Moresi) we present the deployment of the Underworld modelling code onto the grid. All of the StGermain family of codes share their underlying components with Underworld and can be deployed in the same fashion.