We've Moved!
Visit SDSU’s new digital collections website at https://digitalcollections.sdsu.edu
Description
An increasing number of multi-physics modeling applications require the coupling of two or more computational models; which coupling consists of incorporating the interaction between two or more numerical models. One approach to computational coupling is to force elds and variables from one model into the other, without changing the original numerical formulation of either model. This approach is often referred to as "weak" or "loose" coupling. The main objective of this research has been the development of a purely distributed parallel coupler library, called the Distributed Coupling Toolkit (DCT). The DCT is a software library that supports weak coupling between pairs of computational models, whose design provides scalability in both the model's physical complexity and parallel processing. A purely distributed coupling approach reduces inter-process synchronization and improves scalability and resource utilization. This scalability facilitates the implementation of complex model coupling, and enables the inclusion of more model components, higher model resolutions, and dynamic parallel processing con gurations. Performance results show the e ectiveness of the approach adopted with the DCT's implementation. The tests evaluate the distributed approach compared with the commonly used centralized approach. The performance results reveal the drawbacks of a centralized approach and its model coupling implementation, which includes bottlenecks, lack of scalability, and limited model resolution bound by available memory. The distributed approach overcomes all of these problems and restrictions, thereby offering better load balancing. Posterior results of an in-depth performance evaluation of the library implementation lead to a focus on the registration phase (setup) of the DCT. The outcome is a scalable implementation able to handle larger number of processing elements (PEs) without running into memory bounds or PE constraints of the centralized coupling implementations. Coastal simulations require approaches that are able to simulate the effect of the complexity of the bathymetry, rapid currents, and high gradients of density. An approach for use in the General Curvilinear Environmental Model for coastal simulation which features the utilization of the DCT is presented. The seamount test is used to illustrate the potential bene ts of using the DCT to effciently address coastal simulation challenges.
