Out-of-Core LU Decomposition on a Multiple-DSP Platform
Guanglei Wang, Antonello Monti, Gang Quan · 2007
High performance LU decomposition is critical in simulating complex power systems the Virtual Test Bed (VTB) environment. In this paper, we study the parallel LU decomposition problem on the distributed computing platform. We developed a pipelined out-of-core algorithm to reduce the communication cost in the distributed system and implement the algorithm on a commercial multi-DSP board. We also optimized the program code based on the superscalar feature of the target platform to improve the efficiency of this algorithm. The experimental results show that our algorithm can lead to significant improvement over the original algorithm.