GPU Database for Large Geospatial Datasets

Abstract

Nowadays, the approaches and methods of data storage have entirely changed from the traditional SQL-based theorem. The purpose of subsequent data processing has become one of the main aspects. NoSQL databases appeared, which enabled more abstract databases such as graph and time series databases. This paper aims to inspect the benefits of a database principle that works natively on the graphics card, which can be optimal for storing and querying spatial data. This survey introduces the HDF5 file format as a binary data store method, the postGIS system as one of the most widespread GIS data storage methods and the NVIDIA GPUDirect Storage (GDS), which is the high-level concept for enable DMA between the memory of the GPU and with the PCIe devices (e.g. NVME SSD) on Linux. Another objective of the paper is to present which features of these technologies can support the native GPU database concept [7], [8].