Evaluating computational geometry libraries for big spatial data exploration

Abstract

With the rise of big spatial data, many systems were developed on Hadoop, Spark, Storm, Flink, and similar big data systems to handle big spatial data. At the core of all these systems, they use a computational geometry library to represent points, lines, and polygons, and to process them to evaluate spatial predicates and spatial analysis queries. This paper evaluates four computational geometry libraries to assess their suitability for various workloads in big spatial data exploration, namely, GEOS, JTS, Esri Geometry API, and GeoLite. The latter is a library that we built specifically for this paper to test some ideas that are not present in other libraries. For all the four libraries, we evaluate their computational efficiency and memory usage using a combination of micro- and macro-benchmarks on Spark. The paper gives recommendations on how to use these libraries for big spatial data exploration.