Multi-threaded Simulation Software for the JUNO Experiment
Y. Peng · 2025
Read Multi-threaded Simulation Software for the JUNO Experiment by Y. Peng (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-20219.txt.
Introduction
Peidong Yu and co-authors describe simulation software in a 2025 paper for the JUNO neutrino observatory in southern China. The giant underground detector, a sphere of liquid that flashes when neutrinos pass, aims to pin down the ordering of neutrino masses. Faithful simulation is vital, yet the detector's size and energy range strain computing.
The software runs on a shared framework with a standard particle toolkit. Moving from one task at a time to many threads lets modern many-core machines share memory instead of copying it per task. The paper lays out the design of the threaded system and how it fits the experiment's full software suite. Better resource use means more events simulated for the same hardware.
Catalog record
| CREATOR | Y. Peng |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-20219
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Multi-threaded Simulation Software for the JUNO Experiment Y. Peng Abstract The Jiangmen Underground Neutrino Observatory (JUNO) is a multi-purpose experiment under construction in southern China. JUNO aims to determine the neutrino mass ordering and precisely measure the neutrino oscillation parameters by detecting reactor neutrinos from nuclear power plants. In addition to reactor neutrinos, JUNO can study atmospheric neutrinos, solar neutrinos, geo-neutrinos, supernova burst neutrinos, the Diffuse Supernova Neutrino Background (DSNB), and nucleon decay, covering an energy range from keV to TeV. JUNO consists of a 20 kton liquid scintillator detector, equipped with 17,612 20-inch PMTs and 25,600 3-inch PMTs, achieving a 3% energy resolution (at 1 MeV). Precise simulation plays a crucial role in the experiment. Developed using the SNiPER framework and Geant4 toolkit, the JUNO simulation software is a key component of the JUNO offline software (JUNOSW). The large detector size and the broad energy range of interest pose challenges for detector simulation in terms of CPU time and memory consumption. As the computing nodes shift to include multiple integrated CPU cores, traditional single-threaded computing model can lead to large memory footprints and inefficient use of system resources. Multi-threading on many-core architectures can significantly improve resource utilization. This paper introduces the design and implementation of multi-threaded simulation software for the JUNO experiment. 1 Introduction The Jiangmen Underground Neutrino Observotary (JUNO) [ 1 , 2 , 3 ] is a multi-purpose experiment under construction in southern China, expected to finish the construction and start the detector filling by the end of 2024. JUNO is designed to determine the neutrino mass ordering and precisely measure the neutrino oscillation parameters by detecting reactor neutrinos from the Yangjiang and Taishan nuclear power plants. JUNO is also able to observe supernova neutrinos, study the atmospheric neutrinos, solar neutrinos and geo-neutrinos, and perform exotic searches. The JUNO detector is designed with a 20 kton liquid scintillator detector of unprecedented 3% energy resolution (at 1 MeV) at 700 meter deep underground. As shown in Figure 1 , a central detector (CD) contains an acrylic sphere, in which the liquid scintillator is located, surrounded by 17,612 20-inch PMTs and 25,600 3-inch PMTs to detect the scintillation and Cherenkov light with about 78% photocathode coverage. The central detector is immersed in a water pool (WP) instrumented with 2,400 20-inch PMTs to serve as a water Cherenkov detector. The top tracker (TT) is placed on top of the water pool. The water Cherenkov detector and the top tracker form the veto system of JUNO to track and veto muons crossing the detector. Figure 1 : JUNO Detector Offline software is essential for each particle physics experiment in the entire lifecycle of the experiment, from the experiment design, detector construction, to the data analysis. As shown in Figure 2 , the JUNO offline software (JUNOSW) is developed based on the SNiPER [ 4 ] framework, and some other external libraries, such as ROOT [ 5 ] , Geant4 [ 6 , 7 ] , and CLHEP [ 8 ] . The SNiPER framework is a general purpose light-weighted software framework, developed with bi-language C++ and Python. It was originally developed for JUNO, but also adopted by other experiments for example STCF [ 9 ] , HERD [ 10 ] , and nEXO [ 11 ] . Figure 2 : JUNO offline software components JUNO simulation software is a key component of JUNOSW. The simulation software is developed based on Geant4 toolkit, with the simulation event loop fully integrated with the SNiPER. Since rich physics topics can be studied with JUNO, the simulation software can be used to generate different physics events with the energy range from keV to TeV. Full simulation of the primary particles and all the secondary optical photons in JUNO detector is implemented to meet the requirements on detector simulation precision. The simulation software will be introduced in section 2 . Due to the large detector size of JUNO, high energy events can produce tens of million optical photons in the central detector, which pose challenges to detector simulation both in CPU time and memory consumption. Traditional massive Monte Carlo data production is done by having many independent single-threaded jobs running on many CPU cores, where each job runs independently on a single core. However, as CPU clock rates have plateaued, computing nodes have changed to including multiple integrated CPU cores. Traditional single-threaded computing model can lead to large memory footprints and inefficient use of system resources. The memory footprint for JUNO detector simulation jobs will be introduced in section 3 . Multi-threading on many-core architecture can significantly improve utilization of computing resources. The implementation of multi-threaded simulation in JUNOSW will be introduced in section 4 , and the performance results will be introduced in section 5 . Section 6 is the summary. 2 JUNO simulation software JUNO simulation software [ 12 ] is developed based on Geant4, fully taking advantage of elements provided by the SNiPER framework. Figure 3 shows the main components in the SNiPER framework, which consists of Task, Algorithm, Tool and Service components. A Task is an application manager to assemble and control the other components. An Algorithm is a basic unit in data processing. A Tool is a lightweight routine that can be invoked within an Algorithm. A Service is a unit for common functions that can be called anywhere when necessary. As shown in Figure 4 , a simulation framework is implemented with a generator algorithm and a detector simulation algorithm in a task. The generator algorithm calls various generator tools to integrate the different physics generators into the simulation chain. The available generators consist of cosmic muons, reactor neutrinos, atmospheric neutrinos, solar neutrinos, supernova burst neutrinos, DSNB, natural radioactivities, and calibration sources. The detector simulation algorithm invokes a customized Geant4 Run Manager to integrate the underlying Geant4 kernel into the SNiPER. A service named G4Svc manages the run manager, which is invoked during the initialization of the detector simulation algorithm. To configure the geometry, physics list and user action before the initialization of the Geant4 run manager, they are all implemented as tools. Figure 3 : SNiPER components Figure 4 : Detector simulation software 3 Memory footprint and optimization in simulation Since 2012, there’s no obvious increase in CPU clock speeds, but there is a significant increase in the number of processor cores. The memory prices have not decreased significantly, so the ratio of memory to cores tends to decrease. It is now difficult to fully utilize all CPU cores by running multiple jobs on a single computing node, so it is necessary to reduce the amount of memory per core. For high energy events such as muons and atmospheric neutrinos in JUNO simulation, the minimum memory consumption typically exceeds 3 GB. Due to the configuration of computing centers, which provides a fixed amount of memory per CPU core, jobs requiring more memory upon submission will result in the allocation of multiple CPU cores by the computing center to meet the memory requirements. However, only one CPU core is actually utilized to run the job, leaving other cores idle, which results in wasted computational resources and impacts data production for high-energy events. This implies that in large-scale data production for