摘要： 针对集群系统内存不足导致大规模矩阵无法调入的情况，设计并实现一种基于文件拆分与高斯消去的线性方程组求解算法，将大规模系数矩阵文件连续拆分为多个规模可控的子文件进行存储和计算，使服务器空闲内存小且具有较好的计算能力。在ABEEMσπ模型中的应用结果表明，当Altix3700服务器任务提交量大、CPU占有率高时，该算法仍可以计算出蛋白质分子的电荷分布。

关键词: 原子-键电负性均衡模型, 并行计算, 消息传递接口, 高斯消去, 矩阵拆分

Abstract: Aiming at the condition that the limits of cluster memory lead to mass matrix can not import, this paper designs and realizes linear equation solving algorithm based on file splitting and Gaussian elimination. It continuously divides large-scale coefficient matrix files into multiple scale of controllable sub-files to storage and calculation, makes the server free ram small and has the better computation ability. Application result of ABEEMσπ model shows that when Altix3700 server is busy and the CPU share is high, the algorithm can also calculate the charge distribution of protein molecules.

Key words: ABEEMσπ model, parallel computing, Message Passing Interface(MPI), Gaussian elimination, matrix splitting

中图分类号: 

• TP316