A Massively Parallel Implementation of a Numerical Weather Forecast Model

this paper there is hope that the computer power offered by parallel systems can be utilized efficiently for these kinds of applications. Furthermore, this investigation shows that the applied programming model, data-parallelism without explicit message passing, can lead to acceptable efficient code for this application area. This is mainly due to the fact that we deal here with regular computations. For irregular cases, like sparse computations, present day compilers are frequently not smart enough to produce efficient code. The role of the underlying hardware with respect to data-parallelism without explicit message passing is not clear yet. The MasPar implementation is currently ported to the CM5 of Thinking Machines Corporation. Also the T3D of Cray will be investigated. It is expected that the current version can be easily ported to these systems. How efficient the resulting implementations will be is still an open question. Finally, it is interesting that the idea (see section 1) proposed by Richardson far before the invention of what we call computers now, can be realized. Fortunately we do not need thousands of humans to implement it, some sophisticated hardware with accompanying software will do the job. Reference