The investigation of a natural phenomenon explores its occurrence in some space-time reference frame. Typically, scientists select a set of observables that represent the phenomenon and record the latter through a spatio-time series defined by the former. Numerical Simulation reproduces a natural phenomenon by representing the space using a geometry mesh and computing the values of observables through time steps. In this work, we present SimDB, which is a system designed to manage numerical simulation data with observable values varying in space-time. We adopt a multidimensional array data model in which dimensions correspond to space, time and numerical simulation trials. The mapping of numerical simulation data to a multi-dimensional array is, however, not trivial. Particularly, for non-uniform dimensions, as is the case of irregular meshes, a naive mapping produces an extremely sparse array, jeopardising query response time. SimDB proposes technique that transforms nun-uniform dimensions in uniform one, producing compact dense regions, which are efficiently accessed by the multi-dimensional system. SimDB is built on top of SciDB to take advantage of its parallel capacity.
This project is developed in collaboration with the Inria Zenith group, in the context of the Music Project.