多维浮点数据的曲线拟合压缩存储方法

以三维位置信息为代表的多维浮点数据是当前高性能计算机系统处理的主要数据对象，对其进行压缩存储是解决海量多维数据量快速增长的需求与存储空间、存储带宽不足的矛盾的手段之一。现有数据压缩算法对浮点多维数据难以取得理想的压缩效果。将多维浮点数据投影到二维坐标上，使用多项式曲线对其进行拟合，将坐标值的存储转换为多项式系数的存储，实现了多维数据的曲线拟合压缩。算法设计中引入了排序作为数据预处理手段，从而得到了较小的压缩误差。理论分析和实验结果表明，在与现有有损压缩误差相当的条件下，得到了优于现有算法的压缩率。

Curve fitting compression method for massive multi-dimensional floating data storage

Multi dimensional data such as three dimensional position information is one of the major data objects of current high performance computer systems. Its date compression is an important technique to tackle the problem that lack of data storage space and I/O bandwidth cannot meet the demands of rapidly increasing massive multi dimensional data. Existing algorithms are insufficient for multi dimensional floating data compression. A curve fitting method for massive multi dimensional data compression is proposed. Multi dimensional floating point data is projected onto a two dimensional coordinates. By using polynomial curve fitting, the original data is compressed by storing the polynomial coefficients. Sorting is introduced in the design of the algorithm as the data preprocessing means; thereby a smaller compression error is obtained. The theoretical analysis and experimental results show that the compression ratio of our proposed algorithm outperforms the existing algorithms with the same error rate.