A Principled Approach to Distributed Multiscale Computing, from Formalization to Execution

Abstract

In several disciplines, a multiscale approach is being used to model complex natural processes yet a principled background to multiscale modeling is not clear. Additionally, some multiscale models requiring distributed resources to be computed in an acceptable timeframe, while no standard framework for distributed multiscale computing is place. In this paper a principled approach to distributed multiscale computing is taken, formalizing multiscale modeling based on natural processes. Based on these foundations, the Multiscale Modeling Language (MML) is extended as a clear, general, formal, and high-level means to specify scales and interactions in, and as a guide to a uniform approach to crystalize, communicate, develop and execute a multiscale model. With an MML specification, a multiscale model can be analyzed for scheduling or deadlock detection using a task graph. The potential of this method is shown by applying it to two selected applications in nano materials and biophysics.