推进技术 ›› 2021, Vol. 42 ›› Issue (4): 765-775.DOI: 10.13675/j.cnki.tjjs.200724

• 爆震基础机理 • 上一篇    下一篇

斜爆轰波并行数值模拟的化学加速算法性能研究

姜长磊1,董刚1,吴锦涛2   

  1. 1.南京理工大学 瞬态物理重点实验室,江苏 南京 210094;2.北京宇航系统工程研究所,北京 100076
  • 出版日期:2021-04-15 发布日期:2021-04-15
  • 作者简介:姜长磊,硕士生,研究领域为可压缩反应流的数值模拟。E-mail:137868777@qq.com
  • 基金资助:
    航天一院CALT基金项目(CALT2020-ZD01);南京理工大学瞬态物理重点实验室开放基金(6142604180205)。

Performance Study of Chemical Acceleration Algorithm for Parallel Numerical Simulation of Oblique Detonation Wave

  1. 1.Key Laboratory of Transient Physics,Nanjing University of Science and Technology,Nanjing 210094,China;2.Beijing Institute of Astronautical System Engineering,Beijing 100076,China
  • Online:2021-04-15 Published:2021-04-15

摘要: 考虑基元反应条件下的爆轰波精细结构的数值模拟计算量巨大,发展高精度和高效率的计算方法十分必要。以Ma=7的H2/O2/N2预混气来流形成的斜爆轰波为数值模拟对象,研究了一种基于并行计算架构的用于加速化学反应计算的建表算法的计算性能,考察了不同建表策略,即TP(Transposed processing)策略和PLP(Purely local processing)策略,以及不同并行分区数量对算法性能的影响。研究结果显示,本文采用的建表算法能够很好地再现斜爆轰结构,其计算精度不受建表策略和并行计算分区数量的影响;而算法的计算效率则取决于不同分区对应的数据表之间操作的同步性,其中,数据表中节点数据的取回率和数据表设定的尺寸上限都会影响数据表操作的同步性。本文采用的两种建表策略的计算结果表明,TP策略数据表中节点取回率高于PLP策略,故计算效率更高;而计算分区数量越少,则分区对应的数据表尺寸上限越大,数据表的同步性就越好,计算效率也越高。

关键词: 斜爆轰波;建表算法;并行计算;计算性能;数值模拟

Abstract: The numerical simulations of fine detonation wave structures with detailed reaction mechanism are computationally expensive, so it is necessary to develop an algorithm with high-accuracy and high-efficiency. In present study, the numerical simulations of oblique detonation wave generated by a H2/O2/N2 incoming flow with Ma=7 were used to examine computational performance of a tabulation algorithm, which is used to accelerate the computations of chemical reactions on the parallel frame. The effects of the tabulation strategy of algorithm, that is, TP (Transposed processing) and PLP (Puely local processing) strategies, and the parallel zone number on the algorithm performance were focused on. The results show that algorithm in present study can well reproduce the structures of oblique detonation wave, and that the computational accuracy is independent of tabulation strategy and parallel zone number of the algorithm. The computational efficiency of the algorithm is determined by the operation synchronization among the data tables that are mapped into the corresponding computational zones. Both the retrieval ratio of nodes in the tables and the upper limit of table size have the influence on the operation synchronization among the tables. For TP and PLP strategies of the algorithm in present study, the retrieval ratio of nodes in TP strategy is larger than that in PLP strategy and therefore gives rise to the higher computational efficiency. The simulation for case of fewer zone number which has larger upper limit of table size leads to the better synchronization and subsequently the higher computational efficiency.

Key words: Oblique detonation wave;Tabulation algorithm;Parallel computation;Computational performance;Numerical simulation