2016 年 2 巻 p. 16-00119
Numerical simulations such as finite element analyses have increased in size thanks to the explosive development of computer technology. However, large amount of data produced by large scale analysis leads to I/O bottlenecks in simulation, data processing and visualization at the same time. To alleviate this problem, we propose to employ JHPCN-DF technique to large scale finite element analysis for providing a powerful way of data compression. To reduce the quantity of transmission data which is used for further analyses and visualizations, we investigate the possibility of post calculations on local computers and evaluate the precision of post-run results which are calculated by JHPCN-DF encoded data. Over our numerical results, JHPCN-DF performs significantly better than the generic zlib compressor and obtained compression rate of 0.83 with a user defined allowed error ε of 10-2 and 0.68 with ε of 10-4. Our results also indicated that the post-run data calculated by reconstructed data with ε lower than 10-4 achieves almost same level of precision with the one calculated by original data.
