A Layered Communication Optimization Method Based on OpenFOAM

Since Computational Fluid Dynamics (CFD) has emerged as an interdisciplinary science between different subjects, numerous CFD software are increasingly developed on typically layered frameworks. In the literature, only a limited number of optimizations are based on the framework. This paper reveals a general picture of multiphase flow simulation in OpenFOAM and proposes a layered communication optimization method. There are three layers: physical modeling layer, numerical solving layer and parallel computing supporting layer. In these three layers, we respectively eliminate redundant communications in a key module called Multidimensional Universal Limiter for Explicit Solution (MULES), decrease global communications by rearranging preconditioned conjugate gradient (PCG) solver and add non-blocking reduction call interface. Experiments on high-performance computing platforms show that our optimization dramatically enhances the strong scalability of OpenFOAM, increasing the optimal process number by up to three times in both dam-Break flow case and complex 3D head-form case. With a higher degree of parallelism, the shortest execution time, therefore, decreases by 38.16% in the former case and by up to 63.87% in the latter one.