Data Cube Compression with QuantiCubes

Abstract

Data warehouses typically store a multidimensional fact representation of the data that can be used in any type of analysis. Many applications materialize data cubes as multidimensional arrays for fast, direct and random access to values. Those data cubes are used for exploration, with operations such as roll-up, drill-down, slice and dice. The data cubes can become very large, increasing the amount of I/O significantly due to the need to retrieve a large number of cube chunks. The large cost associated with I/O leads to degraded performance. The data cube can be compressed, but traditional compression techniques do not render it queriable, as they compress and decompress reasonably large blocks and have large costs associated with the decompression and indirect access. For this reason they are mostly used for archiving. This paper uses the QuantiCubes compression strategy that replaces the data cube by a smaller representation while maintaining full queriability and random access to values. The results show that the technique produces large improvement in performance.