12-20 July 2017
BEXCO
Asia/Seoul timezone
Home > Timetable > Session details > Contribution details

Contribution

BEXCO - Room F(201/202/203/204)

[GA266] High Performance Computing algorithms for Atmospheric Cherenkov Telescopes

Speakers

  • Thomas VUILLAUME

Primary authors

Co-authors

Description

In the recent years, physics experiments have definitely entered the Big Data era. Future telescopes (such as SKA, the Cherenkov Telescope Array, the LSST, Virgo-LIGO…) will generate more data than ever. This is an exciting time as the analysis and the combination of all these data will lead to new discoveries and even to new ways to do science, but it comes with a price. Data management is now a challenge on its own. The H2020 project ASTERICS is addressing these challenges by bringing together the whole astrophysics and Astroparticle around the ESFRI projects to enable them to interoperate as an integrated, multi-wavelength and multi-messenger telescope. In particular, the OBELICS work package is developing common solutions for the generation, the integration and the analysis of data. Under this framework, new solutions and algorithm based on high performance computing (HPC) to analyze data from Atmospheric Cherenkov Telescopes (IACT) are developed. Several developments are presented in this paper: - an original compression algorithm dedicated to integers. It is therefore especially interesting for physics experiments dealing with digitized signals such as IACT. Coupled with the LZMA algorithm, it considerably reduces the compression time while keeping a maximal compression ratio. - a HPC library with low level algorithms. Applied to the Hillas reconstruction using only reduced momentum and coupled with an adapted data format, these algorithms improve the computing times by a factor greater than 300. - a new reconstruction method based on Single Value Decompositions (SVD). This method compares the data to an image template (generated by Monte Carlo) using a handful of representative values. This considerably reduces the computation time and memory usage while extracting most of the information.