The CSE Software Challenge - Covering the Complete Stack

Typical applications in Computational Science and Engineering (CSE) try to extract information based on a model out of given data, for example to understand physical effects based on numerical simulations. The involved numerical algorithms tend to be complex and their understanding and design require deep mathematical knowledge. Moreover, for physically meaningful results, high accuracy and large-scale problems have to be solved. Writing software for such applications in CSE is a difficult task, since the complex models have to be implemented efficiently on large High Performance Computing (HPC) systems. Therefore, parallel algorithms have to be developed that fit to the underlying parallel hardware, and tools are necessary to analyze and tune the code. In this article we first summarize the demands on CSE software design in general and then present the multi-physics software framework waLBerla as an example for efficient, portable, and productive CSE software.