Estudo das rotinas MPI do Modelo Atmosférico OLAM através de ferramentas de tracing e visualização

Abstract

The weather and climate is an important factor in society as it impacts directly on importante economic and social factors. Predictions are made using atmospheric models that simulate the current weather conditions. To use a considerable amount of data, the execution of this model tends to be slow and could take days. In this context, this paper makes an analysis of the MPI routines implemented in the atmospheric model OLAM (Ocean-Land Atmosphere Model). To this end, we used profiling tools, tracking and viewing MPI routines. As a result, it was examined through load balancing, the routine MPI_Wait had the greatest impact on the processing time of the MPI routines. Represented around 100% in processing time among all MPI routines analyzed. Understanding the operation of the parallel implementation, and the identification of routines that representperformance bottleneck contribute to the improvement of the model.