Abstract: In order to solve the problems of difficulty and low efficiency in traditional parallel programming, this paper presents a parallel programming method based on MapReduce model, realizes matrix parallel LU decomposition under High Performance MapReduce(HPMR) platform. Experimental result shows that the parallel programs implemented via the MapReduce model can meet the need of high-performance numerical computing, and its programming simplicity and readability to enhance the efficiency of parallel programming compared with traditional parallel programming models.

Key words: High Performance MapReduce(HPMR), parallel programming, numerical computation, LU decomposition

摘要： 为了解决传统并行编程难度大、效率低的问题，提出一种基于MapReduce模型的并行编程方法，在高性能MapReduce平台上实现矩阵并行LU分解。实验结果表明，相比传统并行编程模型，MapReduce模型并行程序可较好满足高性能数值计算需求，其编程简洁性和可读性能有效提升并行编程效率。

关键词: 高性能MapReduce, 并行编程, 数值计算, LU分解

CLC Number: 

• TP311.5