NAPA软件的并行化研究和效率分析

Parallel Implementation and Computing Efficiency Analysis for NAPA Code

基于区域分裂思想,以M P I(M essage pass ing in terface)为消息传递库,实现了三维全粘性湍流计算软件NAPA的并行化。对NACA 0012翼型、微型飞行器、高超声速进气道等流场进行了计算和分析,比较了并行和串行软件的计算结果,结果完全一致,并对比了NACA 0012翼型流场的计算结果和实验值,表明NAPA软件的改进是成功的。随后对各个算例的并行效率和加速比进行了对比分析,分析了通信比例、负载平衡度以及通信模式等影响并行效率的几个重要因素。最后使用In te l cluster too ls和V tune工具对NAPA程序进行了算法和代码效率优化,发现了原代码中耗时长的几个子程序并进行了改进,通过对高超声速进气道算例的测试,计算效率提高了55.33%。

The development and the validation of a parallel finite difference Navier-Stokes solver called the NAPA code for unsteady, three-dimensional flow simulations on workstation clusters are described. The solver is parallelized by divided-zone technique and message passing interface（MPI） communication library and validated for flow problems in hypersonic inlet, around micro aviation vehicle and NACA0012 airfoil. Results of the NACA0012 problem by original and the parallelized NAPA code are agreement with the experimental data. All of the comparison show that the parallel implementation of NAPA code is successful. Experiments identify the efficiency and speedups of the parallel code. Results show that the parallel code has good performance. Influencing factors, such as the ratio of communication, load balance and communication mode are discussed. Finally,the code is tuned with Intel Cluster tools and Vtune and the algorithm with the low efficiency in the code is improved. The calculation time for the hypersonic flow is reduced by 55.33%.