圧縮性流体の数値解析における超並列演算処理

概要

論文の詳細を見る
The finite difference method has developed the numerical analysis of compressible flows. In particular, TVD (total variation diminishing) has improved the modeling of shock waves of compressible flows. Numerical results which are in good quantitative agreement well with the experimental results, have never been obtained. One of the reasons is the insufficient resolution for complex reflection of shock wave. Many grid points are necessary to attain the high resolution. Current supercomputers cannot resolve the abovementioned difficulty. Meanwhile, higher calculus processing of more than one teraFLOPS will be required for future numerical computations by LES and DNS.Multiple-instruction multiple-data (MIMD) is considered to be the only type of computer which satisfics the above-mentioned requirement. Difficulty is caused in the development of the application because each processor in MIMD has an independent main memory. In the present paper, the fundamental problem of parallel computation is discussed in numerical analysis of compressive flows using a massively parallel computer (Fujitsu;AP1000). The effect of the method of communication between each processor and the method of dividing the computational area on parallel calculus speed is also discussed.
一般社団法人日本機械学会の論文
1996-07-25