Evaluation and strategy of different data parallel implementation methods of a stiff chemical non-equilibrium flow solver

Abstract

The primary focus of current research concerns three different parallelization implementation methods of a widely used semi-implicit chemically non-equilibrium reactive Euler equations solver on the Vector Parallel Processor (VPP) computer. The main objective is to reduce the relative cost of the convection term calculation and the chemical term computation, and also to optimize the global cost of the numerical simulation by parallel implementation. Three different data parallel methods: two methods based on Domain Decomposition Method (DDM) and one method on Multiblock Structured Algorithm (MSA) are outlined. The characteristics of the three methods are commented on, such as vectorization index, data communication process and parallel I/O. The approaches of the current practical research provide insight into important issues of parallel computing for common applications. Finally, a summary of the parallelization strategy is presented and conclusions are drawn.