Development of High-Performance CFD Coding Method for PC-Cluster Parallel Computing Environment

点击翻译按钮获取中文摘要

Computational fluid dynamics (CFD) is an important component technology for simulation-based design (SBD), which will play essential role in design of new concept ships in next generation. Since beginning of 1990s, large effort has been directed toward development of higher fidelity CFD approach, i.e., Reynolds-averaged Navier-Stokes (RaNS) equation method with free-surface tracking or capturing capability. In general computational mechanics field, the global trend is moving toward utilization of larger mesh, such that computations with over 1 million grid points will be minimum routine task in very near future. The trend will be more accelerated by recent advancement of information technology (IT) especially for low-end high-performance computing facility, e.g., PC-Cluster parallel computing environment. However, availability of computational mechanics codes which fully utilize the parallel capability is still limited, and that holds true for CFD code in ship hydrodynamics. This is du … More e to a fact that parallel coding technique for recent RaNS method has not matured while compiler-based automatic parallel conversion is often found unsatisfactory.This four-year research project (Year-1 through Year-4) concerns development of large-scale high-performance (LSHP) CFD coding method for PC-cluster parallel computing environment. Main objective of the present study is enhancement of capability of RaNS equation solver and its advanced applications for LSHP computing in ship hull form design. In Year-1, focus was placed on setup. and initial evaluation of the present coding environment. The Message Passing Interface (MPI) protocol and the Score-D job scheduling method were implemented in the present system. Detailed performance evaluation of the system was done for high-performance parallel coding of CFD-based nonlinear programming and a domain-decomposition-based commercial RaNS equation solver. In Year-2 through Year-4, primary focus has been on parallel high-performance coding for author's multi-block RaNS method in association with physical domain decomposition, and CFD-based hull form optimization using a parallel-based non-determinative nonlinear programming. These were successfully completed and the results have been presented in more than 10 conference and international journal papers. Less