Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Multi-Level Parallel Pipeline for SPHERE-3 Detector Simulation: From EAS Generation to Image Approximation

V. A. Ivanov, V. I. Galkin, E. A. Bonvech, O. V. Cherkesova, D. V. Chernov, T. A. Kolodkin, N. O. Ovcharenko, D. A. Podgrudkov, T. M. Roganova, M. D. Ziva

TL;DR

A software suite with a multi-step computational pipeline is presented: shower generation in CORSIKA, decoding and cloning of events (C++/OpenMP), ray-tracing of optical photons through the detector model (Geant4 MT), and approximation of images by a lateral distribution function (Python/multiprocessing, iminuit).

Abstract

Optimization of the SPHERE-3 detector configuration, designed to study the mass composition of primary cosmic rays in the energy range 1--1000 PeV by registering Cherenkov light reflected from the snow surface, requires simulation of a large number of extensive air shower events. A software suite with a multi-step computational pipeline is presented: shower generation in CORSIKA, decoding and cloning of events (C++/OpenMP), ray-tracing of optical photons through the detector model (Geant4 MT), and approximation of images by a lateral distribution function (Python/multiprocessing, iminuit). The key property of the problem is its natural atomicity: each event is processed independently at all stages, which provides linear scaling under parallel computation. Thread safety is achieved by architectural means -- shared data are read-only, mutable state is isolated per-worker -- without the use of locks on hot paths.

A Multi-Level Parallel Pipeline for SPHERE-3 Detector Simulation: From EAS Generation to Image Approximation

TL;DR

A software suite with a multi-step computational pipeline is presented: shower generation in CORSIKA, decoding and cloning of events (C++/OpenMP), ray-tracing of optical photons through the detector model (Geant4 MT), and approximation of images by a lateral distribution function (Python/multiprocessing, iminuit).

Abstract

Optimization of the SPHERE-3 detector configuration, designed to study the mass composition of primary cosmic rays in the energy range 1--1000 PeV by registering Cherenkov light reflected from the snow surface, requires simulation of a large number of extensive air shower events. A software suite with a multi-step computational pipeline is presented: shower generation in CORSIKA, decoding and cloning of events (C++/OpenMP), ray-tracing of optical photons through the detector model (Geant4 MT), and approximation of images by a lateral distribution function (Python/multiprocessing, iminuit). The key property of the problem is its natural atomicity: each event is processed independently at all stages, which provides linear scaling under parallel computation. Thread safety is achieved by architectural means -- shared data are read-only, mutable state is isolated per-worker -- without the use of locks on hot paths.
Paper Structure (26 sections, 16 equations, 7 figures, 4 tables)

This paper contains 26 sections, 16 equations, 7 figures, 4 tables.

Table of Contents

  1. Introduction
  2. EAS event generation in CORSIKA
  3. Simulation parameters
  4. Output data format
  5. EAS Event Decoding and Cloning (sim-clone)
  6. Physical Meaning of Cloning
  7. CORSIKA Data Processing
  8. Photoelectron Generation
  9. Parallelism and Data Format
  10. Ray-Tracing in the Detector Model (sim-trace)
  11. Detector Geometry Model
  12. Ray-Tracing Physics
  13. Multithreading Architecture
  14. Shower Image Approximation (sim-fit)
  15. Physical Model
  16. ...and 11 more sections

Figures (7)

  • Figure 1: Computational pipeline for the SPHERE-3 detector simulation. Intermediate data formats and parallelism technologies are indicated for each stage.
  • Figure 2: Cloning scheme: a single Cherenkov light distribution on the snow surface (left) generates multiple images on the detector entrance window at various offsets $(x_{\mathrm{sh}}, y_{\mathrm{sh}})$ within a circle of radius $r_{\mathrm{clone}}$ (right).
  • Figure 3: Geant4 geometric model of the SPHERE-3 detector: spherical mirror, corrector lens, 2653-pixel SiPM mosaic, and protective hood.
  • Figure 4: Multithreading architecture of sim-trace. The master thread creates shared read-only geometry; worker threads dequeue files from the thread-safe FileQueue and write results to isolated output streams.
  • Figure 5: Cherenkov light images on the SPHERE-3 detector mosaic for proton showers (10 PeV) at different distances $R$ from the detector axis. The color scale indicates the number of registered photons per pixel.
  • ...and 2 more figures