CSR-AES: Interactive Parallel Platforms for Multi-Experiment Computational Studies
Scientists and engineers increasingly rely upon computational studies involving multiple experiments (runs of simulation software) in domains ranging from engineering design-space exploration to drug design.
This collaborative project develops foundations for a system that facilitates interactive computational studies involving large numbers of experiments, permitting the user to continuously monitor and steer how the study unfolds, based on partial results. The work specifically targets engineering design applications involving small-to-medium size simulation problems running on tightly-coupled parallel machines, and emphasizes (1) higher-level user control of the overall study; (2) reuse of data from prior experiments in carrying forward new computations; (3) dynamic management of system resources by relying on a tighter coupling between application and system software; and (4) software reuse based on common component architecture (CCA) compliance and standardization of a more permeable system-/solver-level interface.
The system architecture builds upon the SCIRun2 framework and incorporates modules for translating high-level user specifications into required experiments, for allocating parallel machine resources to experiments to optimize study progress, and numerical methods and algorithms for estimating simulation time and resource costs.
The primary impact of the work is expected in engineering design, with broader applicability to additional classes of biomedical and geometric modeling applications. The project will also have important educational applications, by creating hands-on simulated exploration experiences for students, and by training system computer scientists with a deep understanding of specific scientific domains, simulation frameworks, and numerical solver interfaces.
