NEC Corporation Fundamental Research Laboratories
NEC Informatec Systems Ltd
National Aerospace Laboratory
National Aerospace Laboratory
National Aerospace Laboratory
出版者
航空宇宙技術研究所
出版者(英)
National Aerospace Laboratory (NAL)
雑誌名
航空宇宙技術研究所特別資料
雑誌名(英)
Special Publication of National Aerospace Laboratory
巻
30
ページ
13 - 16
発行年
1996-01
会議概要(会議名, 開催地, 会期, 主催者等)
航空宇宙技術研究所 7-9 Jun. 1995 東京 日本
会議概要(会議名, 開催地, 会期, 主催者等)(英)
National Aerospace Laboratory 7-9 Jun. 1995 Tokyo Japan
The Computational Fluid Dynamics (CFD) has an increasingly important role in the design and development of aeronautical and space vehicles. It is well known the costs of the design of aerospace vehicles can be greatly reduced by using the CFD technique. Recent trend of CFD is the computations around more complex and realistic configurations. Considering the limitation of the cost performance of single processor shared memory machine, progress is being made in the direction to the application of Massively Parallel Processor (MPP) architectures. In this presentation, the CFD calculation is introduced using the Message Passing Interface (MPI) in the NAL MPP system, Intel Paragon XP/S25 (336 nodes). Discussions are focused on the MIMD (Multiple Instruction of Multiple Data) type treatment for the Total Variation Diminishing (TVD) scheme code of two dimensional Euler equations. The numerical higher order scheme is the five points difference one. The numerical simulations of shock diffraction around a 90 deg corner are performed. The numerical region is divided into regular small regions which correspond to MPP nodes: each node takes care of local numerical calculations with different boundary conditions. In the present situation, the local region is categorized into nine types. Any small region belongs to one of them. Boundary values with two additional meshes are transferred with each other using MPI in different configurations. Moving wave front are visualized by applying the MPI in the standard MIMD treatment. The efficiency against the space complexity will be discussed by changing mesh size and number of nodes.