The Research On Parallel Algorithm Of Satellite Cloud Motion Winds

Today, with highly developed science and technology, the designing and manufacturing capacity of supercomputers is an important indicator of the science and technology level of a nation and national level, also it is the manifestation of social progress, scientific and technological progress. The parallel computing technology is the cornerstone of the field of high performance computing. With the development of information technology, the information data increase more and more quickly. Facing the vast sea of data, how to use these data to serve human has become an identity of government and scientists. With the powerful computing capacity, parallel computing provides strong technical support for these vast amounts of information and data processing. At present, the parallel computing technology in various fields has played an irreplaceable role, such as weather forecasting, biological science computing, computer simulation, aerospace etc.. With the continuous technological upgrading of meteorological satellite borne equipment, satellite remote sensing data has become more and more abundant, and he challenge comes. How to process these data to produce high density, high precision, high real-time remote sensing products in short time is a serious problems that need to be solved urgently at present.For an important satellite remote sensing products cloud motion wind, this paper proposes three parallel acceleration method of cloud motion winds calculating algorithm based on Open MP parallel framework, GPU parallel framework and MPI parallel framework. Through the analysis of cloud wind serial inversion algorithm, we use divide and rule method to assign each calculating unit some cloud motion wind calculating task according to a certain strategy. Then, all calculating unit begin to deal the tasks assigned. The method can distinctly reduce the time of cloud motion wind calculating. In the algorithm based on MPI parallel framework, a performance prediction method is proposed, which is useful to the MPI cluster load balancing. According to the comparative analysis of experimental data based on the three parallel framework, we find all of the parallel algorithm we proposed can accelerate the efficiency of wind motion wind calculating. The speedup ratio of three parallel has reached 4.96, 112 and 14.96. Among the three parallel algorithm, the GPU acceleration effect is most impressive. In addition, this paper proposes an efficient optimization algorithm for cloud motion wind calculating algorithm. In the case of the minimum loss of the precision of the wind vector, the speedup of the algorithm is up to more than 13.