运用并行算法计算机翼柔度影响系数矩阵

Computation of Wing Flexibility Coefficient Matrix with Parallel Algorithm

并行算法是一种借助数学方法和程序设计在并行计算机上高效处理数据的算法,它是发挥并行计算机处理数据能力强的功能的必要保证.超级计算机必须借助并行算法才能达到所标称的运算速度.并行计算可分为同步并行计算和异步并行计算.与之相适应的分别为同步并行算法和异步并行算法.两种并行算法的选择取决于所用并行机的体系结构.在 VAX-780机上计算某飞机机翼柔度影响系数矩阵(简称柔阵)需要10hCPU 机时.作者采取了—些并行化措施,成功地在中国超级计算机——银河-Ⅰ上仅用7min30sCPU 机时作同样计算.该机翼结构的5300阶刚度矩阵 K 已作好三角分解,选出160点(点号的集合为 L),需要导出这些点的柔度影响系数矩阵 F_L(160×160),即在每一点上加 Y 向单位载荷 i_K(K∈L)。

The authors take certain parallelism measures and succeed in computing the flexibility coefficient matrix in 7 minutes 30 seconds on Chinese supercomputer YH-1. The measures taken by the authors to improve computation efficiency are: (1) The data structure of the wing structure stiffness matrix in the internal memory is changed. This measure reduces greatly the search time needed for recalling internal memory. It also eliminates possible appearance of so called data flow conflict phenomenon. (2) The huge matrix is resolved into blocks and the authors' algorithm makes the exchange between the internal memory and the external memory possible and convenient.