Bandlimited OLAP cubes for interactive big data visualization

Caleb Reach, Chris North

Abstract

Visualizations backed by data cubes can scale to massive datasets while remaining interactive. However, the use of data cubes introduces artifacts, causing these visualizations to appear noisy at best and deceptive at worst. Moreover, data cubes highly constrain the space of possible visualizations. For example, a histogram backed by a data cube is constrained to have a bin width that is a multiple of the data cube bin size. Similarly, for dynamic queries backed by data cubes, query extents must be aligned with bin boundaries. We present bandlimited OLAP (online analytical processing) cubes (BLOCs), a technique that uses established tools from digital signal processing to generate interactive visualizations of very large datasets. Based on kernel density plots and Gaussian filtering, BLOCs suppress the artifacts that occur in data cubes and allow for a continuous range of zoom/pan positions and continuous dynamic queries.