Aachen 2019 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz

AKPIK 3: Machine-learning methods and computing in astroparticle physics

AKPIK 3.10: Vortrag

Mittwoch, 27. März 2019, 17:30–17:40, H06

Highly parallel CORSIKA processing — •Dominik Baack — TU Dortmund, Dortmund, Germany

The increasing amount of simulation necessary for an effective analysis in the high energy regime requires new approaches for the optimization of the existing simulation software like CORSIKA.

The dominating factor for the simulations including Cherenkov photons is the creation and propagation of those. With more than half of the overall runtime and billions of executions for each shower this part benefits heavily from any possible optimization. The approach taken for the acceleration of this part is the parallel implementation in OpenCL. This can be used on CPU or even faster on specialized hardware like GPUs.

The second approach focuses on the distribution of a single simulation over multiple independent nodes over the network with the ZeroMQ library. This allows the parallel work of multiple nodes of different architecture on a single shower or the control of multiple independent showers via a single control instance. Optimization focusing on low power usage allows the distribution over hundreds of low power ARM cores via a central computing node.

The methods used and first results of different experiments done will be shown.