| Popis: |
The ecological condition of the Continental Shelf is of great concern for many countries. The understanding of the integrated effects that result in the present and future situation requires answers on many research topics, such as flow modelling, including turbulence and large eddy simulation, transport processes, chemistry, ecology, etc. This paper focuses on 'high performance computing issues' for flux modelling. Flux modelling of contaminants, nutrients and ecosystem parameters in general in the Continental Shelf is essential to improve the understanding of this important ecosystem. From a computational point of view, flux modelling is a 'grand challenge'; its computational demands are so huge that the present state-of-the-art does not yield numerical models of desired accuracy. Several assumptions and simplifications are needed to arrive at 'manageable' numerical models whose run times are acceptably low. In an attempt to relieve the computational burden on flux modelling, a significant amount of research at the institutes participating in the NOWESP project has been and still is directed towards high performance computing techniques. In the quest for faster models several approaches are used: * parallelization of (sequential) codes on parallel/vector computers of shared memory type; * parallelization of (sequential) codes on parallel computers with distributed memory; * development of numerical techniques that (are expected to) lead to better efficiency and robustness of flux models. Developments of the latter kind usually take into account the use of a parallel machine but it also includes the implementation of sophisticated iterative solution techniques for solving systems of equations. This may already pay off on sequential machines. In this paper we assess the progress made in recent years on high performance flux modelling. In the spirit of the learning-by-doing process adopted in the NOWESP project, this assessment is followed by some recommendations for future research to further overcome the lack of computational power that is needed for full scale flux models. |
