针对在实际应用中超出内存容量的大规模矩阵不可调入的问题,设计并实现了一种面向A矩阵连续划分的、子文件大小可调节的求解三角形方程组的并行方案。理论上只要在不超过硬盘容量的前提下,该算法可以在小规模内存条件下求解任意规模的矩阵。最后给出在Altix3700服务器上基于ABEEMαπ模型的数值实验,实验结果表明了该算法的有效性。
A parallel algorithm is proposed, in which it' s A matrix is divided into several adjustable smaller matrix files to solve the problem that the matrix' s scale is larger than the memory' s capacity. Theoretically, this algorithm may solve the random scale matrix under the small condition of memory as long as it does not surpass the hard disk' s capacity. Finally, numerical experiments about the algorithm based on ABEEMαπ model on Altix3700 are carried on, and it obtains the correct results and confirms the algorithm' s validity.