A practical multiple processor programming model for various distributed memory architectures

Recent advances in developing parallel algorithms that can be ported to various architectures have emerged. We have developed a methodology that can be used to easily port parallel algorithms to smaller systems such as commercial processor systems or to larger systems such as message passing super computers. This flexibility is important because it leads to scalable algorithm/architecture pairs. The methodology has been. successfully used to map parallel algorithms to a system containing multiple digital signal processors and to the GRAY T3E. Each proved to be scalable up and down and had high performance as compared to the processor ratings. We tested the methodology with the Kalman filter as used for data assimilation. This application has high computational intensity when used for the atmospheric forecast model which we also implemented.