D8-tree: a de-normalized approach for multidimensional data analysis on key-value databases

Abstract

The shift to more parallel and distributed computer architectures changed how data is managed consequently giving birth to a new generation of software products, namely NoSQL. These products offer a scalable and reliable solution for "Big Data", but none of them solves the problem of analyzing and visualizing multidimensional data. There are solutions for scaling analytic workloads, for creating distributed databases and for indexing multidimensional data, but there is no single solution that points to all three goals together. We propose the D8-tree, a De-normalized Octa-tree index that supports all three goals. It works with both analytical and data-thinning queries on multidimensional data ensuring, at the same time, low latency and a linear scalability. We have implemented a D8-tree prototype, and we compared it with PostGIS on a set of queries required by an in-house HPC application. We found the D8-tree outperforming PostGIS in all tested queries, with a performance gain up to 47 times.