Poster Abstract

P7.1 Johan Bregeon (CNRS IN2P3 LPSC)

Theme: Data discovery across heterogeneous datasets

Optimisation of the air shower simulation software in the current CTA production setup

The Cherenkov Telescope Array Observatory has started to build the world largest ground-based gamma-ray observatory, with open time offered for allocation to the astronomical community. The instrument responses are derived from Monte Carlo simulations of the full chain, starting from the extensive air shower in the atmosphere via the Corsika software, going through Cherenkov light production and transmission down to the telescope optics and camera electronics. Such simulations are very demanding both in term of processing and storage resources: the CTA consortium has been using an average of one hundred millions normalized cpu each year in the past 5 years for that purpose. Optimizing the cpu consumption of simulations is hence of primary importance both to reduce infrastructure costs and the environmental imprint of the CTA.

We present here both the methodology and results of in-depth profiling and optimization procedure applied on the code used to produce and propagate through the atmosphere the Cherenkov light of air showers, as set up in the current CTA production environment. The main steps of the work comprise profiling, code review and rationalization, algorithm optimization and eventually code vectorization via the use of Single Instruction Multiple Data (SIMD) instructions obtained through a proper set of compiler options. Perspectives for further optimizations based both on work on memory pattern and on the possible reduction of the precision of targeted computations will also be presented.