École Nationale Supérieure de
Techniques Avancées

Abstract

With the advance of high performance computing, the results of numerical simulations get bigger and bigger. The cost and performances to write data onto disks are then a problem when it comes to process them. In order to avoid to write onto disks, on-going methods propose to allocate a few more resources to do simulation and treatments simultaneously. The available memory for the treatment depending on the number of allocated resources, all the data will not fit entirely in memory when they get bigger. During this internship, streaming is studied to allow on-going analyze of the data of the simulation. In scientific visualization, visualization pipelines are applied to the data to create a graphical rendering. Splitting the data into pieces and applying the pipeline piece by piece is a way to reduce the memory footprint. In this internship, two methods of streaming are proposed for VTK, a visualization library well-known among the community. The methods are first analyzed with respect to their memory footprints on one core and are then adapted to run on several cores,using ParaView, the parallel extension of VTK. Performance analyses are then conducted and the best method is adapted to propose a first step towards on-going treatments of a molecular dynamic simulation.