Poster Abstract

P9.13 Ugo Becciani (INAF - Istituto Nazionale di Astrofisica)

Theme: Evolution of software development and management

Big data analysis and reduction on Exascale infrastructure: the AVU-GSR application in the EuroExa p

The goal of Gaia AVU-GSR application is to produce a Global Sphere Reconstruction using a subset of the Gaia observations adopting a modified version of the PPN-RAMOD model.
The problem consists of a large and sparse system of equations of the order up to 10^10 * 10^8. To solve such a system, we use a hybrid implementation of the iterative PC-LSQR with the aid of some parallelization techniques and of an ad-hoc compression algorithm of the sparse system matrix. AVU-GSR code is currently used in production for the Gaia Mission running on Cineca Tier0 and is one of the application selected in the Horizon-2020 funded EuroExa project for the system-level co-design approach to develop a HPC compute node. These nodes will balance general purpose compute and reconfigurable acceleration for peta-flop, peta-byte and tera-bit levels of compute, storage and communication respectively that are capable of addressing the resource demands across the range of targeted HPC application codes. We present the preliminary results we obtained using the new paradigm OmpSs with MPI, and the future perspectives that include the usage of accelerators like GPU and FPGA.