Enabling Visualization of Massive Datasets Through MPP Database Architecture

Abstract

We are developing a novel visualization architecture which is specifically designed to render very large (terabyte scale) datasets. Our method differs from the classic visualization pipeline of Harber and McNabb. In particular we eliminate the need to create geometric objects, for example surfaces composed of polygons, as a stage before rendering. Such objects require specialist HPC servers for their creation and manipulation; our solution eliminates the need for such servers. We replace the geometric objects by structures stored and tagged in a database next to the original dataset; we call these Spatially Registered Data Structures (SRDS). Such structures are linked to a single rendering pipeline through the on-the-fly creation of a Feature Embedded Spatial Volume (FESVo). This solution exploits recently developed capabilities of in-database Massive Parallel Processing (MPP) and parallel data streaming, together with the rapidly developing capabilities of GPUs. We describe an early prototype of an architecture applied to seismic data from the oil and gas industry.