Accessible Parallelization for the Open Modeling Interface

As the availability of computing infrastructure continues to increase, so too does the need for accessible means for utilizing those resources. An effective approach is to enable desktop-oriented scientific software tools and frameworks to support execution on high performance cyberinfrastructure in a way that is transparent to the user. We have found this to be the case in our ongoing environmental modeling study in which we are applying multidisciplinary, integrated models to the study of a depleting aquifer. Our models are linked together using the Open Modeling Interface (OpenMI) which provides a composition framework for the sequential execution of model components. In this work we investigate the potential for incorporating parallelism into the OpenMI as a first-class citizen. We present a general solution in which model components may be executed in parallel without requiring changes to their source code. An alternate solution achieves greater parallelism through simultaneous invocations of individual components, but requires them to be modified in some cases. These can result in significant reductions in simulation runtimes on both multi-core desktop machines as well as in high performance computing environments. We demonstrate this potential speedup in a performance study in which the application of the general solution achieved 86% of linear speedup when executed on a high performance machine with 80 cores.