基于预测的纯量场数据压缩技术研究

Predictor Based Fast and Efficient Compression of Scalar Fields

大规模数值模拟产生了海量数据,对数据存储空间和I/O带宽都形成了挑战。针对纯量场数据,研究了外插与内插预测算子,分析了两种预测算子的优缺点,并提出了基于外插预测的纯量场无损压缩方法以及基于内插预测的纯量场有损压缩方法。提出的压缩方法的突出优点是内存开销比较小,适合于大规模的纯量场数据的压缩。使用光滑数学模型数据和真实物理模拟数据进行的测试实验表明,提出的基于预测的纯量场压缩方法取得了良好效果。

Data sets from scientific simulation are growing in size at an ever-increasing pace. We presented a comparative analysis of extrapolating and interpolating predictors. Then we proposed two simple methods for lossless or lossy compressing large and regularly sampled scalar fields. Our method is particularly attractive for its small usage of memory and suitable for extremely large data sets. Experiments using smooth mathematical model data and true plasma physics simulation data indicate that our method performs well and encouraging.