Section: New Results

Distributed Systems

The work carried in previous years on distributed High-Performance Computing for e-Science workflows has enlightened the need for appropriate tools and methods to manage petabyte and exabyte volumes of data. This has been the focus of the work carried by Benoit Lange during his Post-Doc position in 2014. It was dedicated to the definition and prototyping of a large-scale HPC platform to support the execution of application solvers, efficient storage and management of large-volumes of data produced by the simulation applications and the visualization of their results on high-end graphics workstations. This platform also includes analytics software to produce specific results corresponding to the user queries. This is based on the Hadoop ecosystem [59] . Is is central for the communication between the dedicated HPC nodes running the solvers and the visualization nodes interfacing the end-users. It includes high-speed storage with dedicated file systems on specific nodes, and long-term storage for reference data using magnetic juke-boxes that store petabytes of application data. This work is supported by an FP7 project in which Inria is responsible for the Data Management work-package (Call FP7-2013-ICT-11, Grant 619439, 2014-2016). The partners of the project, named VELaSSCo (Visualization for Extremely Large Scale Scientific Computing), are : CIMNE (SP, coordinator), JOTNE and SINTEF (No), ATOS (SP), Fraunhofer IGD (D) and the University of Edinburg (UK).