Abstract Interpretation: Testing at Scale without Testing at Scale

In scientific computing, scaling issues frequently occur as more processors are utilized and/or the problem size is increased. Applications that run well on small problems and small machines, don\´t necessarily work well, or at all when executed at larger scale. It is desirable to avoid using expensive supercomputer time for discovering and correcting problems; instead, we would like to do this beforehand using analysis and/or testing.In this position paper, we discuss testing and problem and target platform specific static analyses of scientific programs developed using the Super Instruction Architecture, a parallel programming environment for computational chemistry. In particular, a set of tools which perform abstract interpretation, allow us to answer questions about resource usage and computation time for specific "scaled-up" inputs and machine configurations without needing to run the full program on the target platform.