Parallel implementation of large-scale CFD data compression toward aeroacoustic analysis

A data compression method was developed for large-scale, unsteady flow simulation data. The method includes three steps of discrete wavelet transform, quantization and entropy encoding. The discrete wavelet transform extracts only important features in the flow field from the entire simulation data. In the quantization process, high quantization bit rate is assigned only to the important region extracted by the previous transform. Finally, entropy encoding reduces the quantized data size. This data compression method was implemented to unsteady incompressible flow simulation data obtained by a block-structured Cartesian mesh method named Building-Cube method. The framework of the Building-Cube method enables not only easy implementation of data compression but also highly efficient parallelization by a very simple scheme for load balance. Numerical test cases demonstrated that the present compression method gave a high compression ratio with good quality of compressed data in terms of velocity distributions, turbulent statistics, and aeroacoustic analysis, as well as high scalability in parallelization.