Efficient parallel implementation of large scale 3D structured grid CFD applications on the Tianhe-1A supercomputer

In this paper, a parallelization for a large scale CFD application with mixed one-to-one multiblock/overset structured grid was implemented into our in-house TH-CFD code running on Tianhe-1A supercomputer system. Strategies at multiple software levels were employed in a mutually supportive way for overall performance enhancement, and they include grid repartition, MPI+OpenMP hybrid programming, use of multiple programming languages adaptively for different software components, and adaptive thread scheduling for load balancing. Numerical testing was carried out and the testing results have shown high efficiency of our parallel implementation, supporting the effectiveness of our parallelization strategies.