A Decomposition-Based Partition Method For Matrix-Multiplication In The Mapreduce Model

Matrix multiplication is widely used in various applications of linear algebra. The efficiency of matrix multiplication in the MapReduce model is bounded by the workload of intermediate key-value pairs and transmission cost between the map phase and the reduce phase. An appropriate partition of matrix data can reduce unnecessary transmission cost of the two phases. A matrix decomposition based method is proposed in this paper to reduce the transmission cost and provides effective matrix multiplication computation. The experimental result shows that the proposed method can lower the transmission cost as the matrix size increases.