Spatial Indexing of Large Multidimensional Databases

Abstract: Scientific endeavors such as large astronomical surveys generate databases on the terabyte scale. These, usually multidimensional databases must be visualized and mined in order to find interesting objects or to extract meaningful and qualitatively new relationships. Many statistical algorithms required for these tasks run reasonably fast when operating on small sets of in-memory data, but take noticeable performance hits when operating on large databases that do not fit into memory. We utilize new software technologies to develop and evaluate fast multidimensional indexing schemes that inherently follow the underlying, highly non-uniform distribution of the data: they are layered uniform grid indices, hierarchical binary space partitioning, and sampled flat Voronoi tessellation of the data. Our working database is the 5-dimensional magnitude space of the Sloan Digital Sky Survey with more than 270 million data points, where we show that these techniques can dramatically speed up data mining operations such as finding similar objects by example, classifying objects or comparing extensive simulation sets with observations. We are also developing tools to interact with the multidimensional database and visualize the data at multiple resolutions in an adaptive manner.