Big Data architecture for large-scale scientific computing

By B. Lange and T. Nguyen (INRIA)

Abstract : Today, the scientific community uses massively simulations to test their theories and to understand physical phenomena. Simulation is however limited by two important factors: the number of elements used and the number of time-steps which are computed and stored. Both limits are constrained by hardware capabilities (computation nodes and/or storage). From this observation arises the VELaSSCo project1. The goal is to design, implement and deploy a platform to store data for DEM (Discrete Element Method) and FEM (Finite Element Method) simulations. These simulations can produce huge amounts of data regarding to the number of elements (particles in DEM) which are computed, and also regarding to the number of time-steps processed. The VELaSSCo platform solves this problem by providing a framework fulfilling the application needs and running on any available hardware. This platform is composed of different software modules: a Hadoop distribution and some specific plug-ins. The plug- ins which are designed deal with the data produced by the simulations. The output of the platform is designed to fit with requirements of available visualization software.