Development of a spatial domain decomposition scheme for Monte Carlo neutron transport

Manuel García, Diego Ferraro, Victor Hugo Sanchez-Espinoza, Luigi Mercatali, Jaakko Leppänen, Ville Valtavirta

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingsScientificpeer-review

Abstract

The computing power available nowadays to the average Monte-Carlo-code user is sufficient to perform large-scale neutron transport simulations, such as full-core burnup or high-fidelity multiphysics. In practice however, software limitations in the majority of the available Monte Carlo codes result in a low efficiency when running in High Performance Computing (HPC) environments, the main issues being inadequate memory utilization and poor scalability. The traditional parallel processing scheme based of splitting particle histories among processes requires domain replication across nodes, and therefore the memory demand for each computing node does not scale, and a memory bottleneck appears for large-scale problems. The scalability of this approach usually limits the resources that can be used efficiently to a small number of nodes/processors. Consequently, massively parallel execution is not viable with particle-based parallelism, at least not by itself. In this work we propose a Spatial Domain Decomposition (SDD) approach to develop an efficient and scalable Monte Carlo neutron transport algorithm. Breaking down the geometry into subdomains, a distributed memory scheme can be used to reduce the in-node memory demand, allowing the simulation of large-scale memory-intensive problems. Additionally, with an efficient neutron tracking algorithm the overall speedup can be significantly improved.

Original languageEnglish
Title of host publication26th International Conference on Nuclear Engineering
Subtitle of host publicationNuclear Fuel and Material, Reactor Physics, and Transport Theory
PublisherAmerican Society of Mechanical Engineers ASME
Number of pages7
Volume3
ISBN (Print)978-0-7918-5145-6
DOIs
Publication statusPublished - Nov 2018
MoE publication typeNot Eligible
Event26th International Conference on Nuclear Engineering, ICONE 2018 - London, United Kingdom
Duration: 22 Jul 201826 Jul 2018
Conference number: 26

Conference

Conference26th International Conference on Nuclear Engineering, ICONE 2018
Abbreviated titleICONE 2018
CountryUnited Kingdom
CityLondon
Period22/07/1826/07/18

Fingerprint

Neutrons
Decomposition
Data storage equipment
Scalability
Geometry
Processing

Cite this

García, M., Ferraro, D., Sanchez-Espinoza, V. H., Mercatali, L., Leppänen, J., & Valtavirta, V. (2018). Development of a spatial domain decomposition scheme for Monte Carlo neutron transport. In 26th International Conference on Nuclear Engineering: Nuclear Fuel and Material, Reactor Physics, and Transport Theory (Vol. 3). [ICONE26-82144] American Society of Mechanical Engineers ASME. https://doi.org/10.1115/ICONE26-82144
García, Manuel ; Ferraro, Diego ; Sanchez-Espinoza, Victor Hugo ; Mercatali, Luigi ; Leppänen, Jaakko ; Valtavirta, Ville. / Development of a spatial domain decomposition scheme for Monte Carlo neutron transport. 26th International Conference on Nuclear Engineering: Nuclear Fuel and Material, Reactor Physics, and Transport Theory. Vol. 3 American Society of Mechanical Engineers ASME, 2018.
@inproceedings{390a0550f8ee4b428e3cf64ba3afd0b3,
title = "Development of a spatial domain decomposition scheme for Monte Carlo neutron transport",
abstract = "The computing power available nowadays to the average Monte-Carlo-code user is sufficient to perform large-scale neutron transport simulations, such as full-core burnup or high-fidelity multiphysics. In practice however, software limitations in the majority of the available Monte Carlo codes result in a low efficiency when running in High Performance Computing (HPC) environments, the main issues being inadequate memory utilization and poor scalability. The traditional parallel processing scheme based of splitting particle histories among processes requires domain replication across nodes, and therefore the memory demand for each computing node does not scale, and a memory bottleneck appears for large-scale problems. The scalability of this approach usually limits the resources that can be used efficiently to a small number of nodes/processors. Consequently, massively parallel execution is not viable with particle-based parallelism, at least not by itself. In this work we propose a Spatial Domain Decomposition (SDD) approach to develop an efficient and scalable Monte Carlo neutron transport algorithm. Breaking down the geometry into subdomains, a distributed memory scheme can be used to reduce the in-node memory demand, allowing the simulation of large-scale memory-intensive problems. Additionally, with an efficient neutron tracking algorithm the overall speedup can be significantly improved.",
author = "Manuel Garc{\'i}a and Diego Ferraro and Sanchez-Espinoza, {Victor Hugo} and Luigi Mercatali and Jaakko Lepp{\"a}nen and Ville Valtavirta",
year = "2018",
month = "11",
doi = "10.1115/ICONE26-82144",
language = "English",
isbn = "978-0-7918-5145-6",
volume = "3",
booktitle = "26th International Conference on Nuclear Engineering",
publisher = "American Society of Mechanical Engineers ASME",
address = "United States",

}

García, M, Ferraro, D, Sanchez-Espinoza, VH, Mercatali, L, Leppänen, J & Valtavirta, V 2018, Development of a spatial domain decomposition scheme for Monte Carlo neutron transport. in 26th International Conference on Nuclear Engineering: Nuclear Fuel and Material, Reactor Physics, and Transport Theory. vol. 3, ICONE26-82144, American Society of Mechanical Engineers ASME, 26th International Conference on Nuclear Engineering, ICONE 2018, London, United Kingdom, 22/07/18. https://doi.org/10.1115/ICONE26-82144

Development of a spatial domain decomposition scheme for Monte Carlo neutron transport. / García, Manuel; Ferraro, Diego; Sanchez-Espinoza, Victor Hugo; Mercatali, Luigi; Leppänen, Jaakko; Valtavirta, Ville.

26th International Conference on Nuclear Engineering: Nuclear Fuel and Material, Reactor Physics, and Transport Theory. Vol. 3 American Society of Mechanical Engineers ASME, 2018. ICONE26-82144.

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingsScientificpeer-review

TY - GEN

T1 - Development of a spatial domain decomposition scheme for Monte Carlo neutron transport

AU - García, Manuel

AU - Ferraro, Diego

AU - Sanchez-Espinoza, Victor Hugo

AU - Mercatali, Luigi

AU - Leppänen, Jaakko

AU - Valtavirta, Ville

PY - 2018/11

Y1 - 2018/11

N2 - The computing power available nowadays to the average Monte-Carlo-code user is sufficient to perform large-scale neutron transport simulations, such as full-core burnup or high-fidelity multiphysics. In practice however, software limitations in the majority of the available Monte Carlo codes result in a low efficiency when running in High Performance Computing (HPC) environments, the main issues being inadequate memory utilization and poor scalability. The traditional parallel processing scheme based of splitting particle histories among processes requires domain replication across nodes, and therefore the memory demand for each computing node does not scale, and a memory bottleneck appears for large-scale problems. The scalability of this approach usually limits the resources that can be used efficiently to a small number of nodes/processors. Consequently, massively parallel execution is not viable with particle-based parallelism, at least not by itself. In this work we propose a Spatial Domain Decomposition (SDD) approach to develop an efficient and scalable Monte Carlo neutron transport algorithm. Breaking down the geometry into subdomains, a distributed memory scheme can be used to reduce the in-node memory demand, allowing the simulation of large-scale memory-intensive problems. Additionally, with an efficient neutron tracking algorithm the overall speedup can be significantly improved.

AB - The computing power available nowadays to the average Monte-Carlo-code user is sufficient to perform large-scale neutron transport simulations, such as full-core burnup or high-fidelity multiphysics. In practice however, software limitations in the majority of the available Monte Carlo codes result in a low efficiency when running in High Performance Computing (HPC) environments, the main issues being inadequate memory utilization and poor scalability. The traditional parallel processing scheme based of splitting particle histories among processes requires domain replication across nodes, and therefore the memory demand for each computing node does not scale, and a memory bottleneck appears for large-scale problems. The scalability of this approach usually limits the resources that can be used efficiently to a small number of nodes/processors. Consequently, massively parallel execution is not viable with particle-based parallelism, at least not by itself. In this work we propose a Spatial Domain Decomposition (SDD) approach to develop an efficient and scalable Monte Carlo neutron transport algorithm. Breaking down the geometry into subdomains, a distributed memory scheme can be used to reduce the in-node memory demand, allowing the simulation of large-scale memory-intensive problems. Additionally, with an efficient neutron tracking algorithm the overall speedup can be significantly improved.

UR - http://www.scopus.com/inward/record.url?scp=85056162848&partnerID=8YFLogxK

U2 - 10.1115/ICONE26-82144

DO - 10.1115/ICONE26-82144

M3 - Conference article in proceedings

AN - SCOPUS:85056162848

SN - 978-0-7918-5145-6

VL - 3

BT - 26th International Conference on Nuclear Engineering

PB - American Society of Mechanical Engineers ASME

ER -

García M, Ferraro D, Sanchez-Espinoza VH, Mercatali L, Leppänen J, Valtavirta V. Development of a spatial domain decomposition scheme for Monte Carlo neutron transport. In 26th International Conference on Nuclear Engineering: Nuclear Fuel and Material, Reactor Physics, and Transport Theory. Vol. 3. American Society of Mechanical Engineers ASME. 2018. ICONE26-82144 https://doi.org/10.1115/ICONE26-82144