1. Memo
(1) The goal of this paper
The goal of this paper is to accelarate visualization and quantitive analysis of simulation data.
(2) What simulation data is
A simulation is forecasting or estimating numerical value as velocity and pressure etc. at each 3D space corrdinate and each time in area such as astronomy; medicine; biology. A simulation solves a set of differential equations at an initial condition through discretization of the space. A set of simulation data are point clouds using several initial conditions of differential equation. Size of simulation data may become terabytes.
(3) What quantitive analysis of simulation data is
An example of quantitive analysis is aggregation of distance between simulation value and reference value.
(4) Challenge
Row store is inadequate for analytical query. Array DBMS such as SciDB can intuitive represent simulation data because a set of simulation data is multidimensional array. However a naive managing approach produce sparse array so cannnot accelarate analytical queries when using discretization of the space with irregular meshes.
(5) Approach
When simulation data is fit in memory: managing data using in-memory column store such as MonetDB. Otherwise mapping simulation data to evenly distributed storage blocks using equi-depth histogram and space filling curves.
2. References
[1] H. Lustosa, F. Porto, P. Blanco, Database System Support of Simulation Data, PVLDB2016.
https://www.vldb.org/pvldb/vol9/p1329-lustosa.pdf