Abstract:
The technique for high-performance computing cluster architecture using messaging passing interface (MPI) is presented. The technique is based on the spatial domain decomposition. It is shown that when calculating multistage each spatial subregion separate process can achieve a multiple number of computational processes linear growth performance multiprocessor applications up to the start time / end of the computational process. An assessment of the minimum amount of computational grid points attributable to each computational process, at which the linear increase in performance.
References