A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Shared memory parallelization for high-fidelity large-scale 3D polyhedral particle simulations
Abstract Particle shape plays a vital role in granular material behavior, but simulations with realistic particle shapes are uncommon due to significant computational demands of complex particle geometry representation. In this work, BLOKS3D, a polyhedral Discrete Element Method (DEM) and impulse-based DEM (iDEM) codes are parallelized to enable large-scale simulations with realistic particle shapes on readily accessible multi-core machines. Data structures used in the original codes were redesigned and optimized, leading to 15% improved performance of the original serial codes. New parallel algorithms were developed resulting in 28 times performance improvement on a 48-core (quad-CPU) shared memory system over single core serial algorithm. The parallelized 3D polyhedral DEM and iDEM were applied to series of column collapse simulations. The codes successfully reproduced the runout distance in granular column collapse experiments. The particle force data from both parallelized DEM and iDEM matched data from the serial algorithm. The new parallel implementation of iDEM was then demonstrated with unprecedented 52 million 3D polyhedral particles simulations. This work will benefit future granular material studies with the newly introduced capacity to run large-scale simulations with realistic particle shapes on shared memory hardware platforms readily accessible to many engineers and researchers.
Shared memory parallelization for high-fidelity large-scale 3D polyhedral particle simulations
Abstract Particle shape plays a vital role in granular material behavior, but simulations with realistic particle shapes are uncommon due to significant computational demands of complex particle geometry representation. In this work, BLOKS3D, a polyhedral Discrete Element Method (DEM) and impulse-based DEM (iDEM) codes are parallelized to enable large-scale simulations with realistic particle shapes on readily accessible multi-core machines. Data structures used in the original codes were redesigned and optimized, leading to 15% improved performance of the original serial codes. New parallel algorithms were developed resulting in 28 times performance improvement on a 48-core (quad-CPU) shared memory system over single core serial algorithm. The parallelized 3D polyhedral DEM and iDEM were applied to series of column collapse simulations. The codes successfully reproduced the runout distance in granular column collapse experiments. The particle force data from both parallelized DEM and iDEM matched data from the serial algorithm. The new parallel implementation of iDEM was then demonstrated with unprecedented 52 million 3D polyhedral particles simulations. This work will benefit future granular material studies with the newly introduced capacity to run large-scale simulations with realistic particle shapes on shared memory hardware platforms readily accessible to many engineers and researchers.
Shared memory parallelization for high-fidelity large-scale 3D polyhedral particle simulations
Park, Eun Hyun (author) / Kindratenko, Volodymyr (author) / Hashash, Youssef M.A. (author)
2021-01-01
Article (Journal)
Electronic Resource
English
Decomposition and Parallelization of Large Scale Framed Structure Optimization
| British Library Conference Proceedings | 1995
Large Scale Numerical Simulation via Parallelization and Reconfigurable Computing Hardware
| British Library Conference Proceedings | 2006
Parallelization of the Particle Model SIMPAR
| British Library Conference Proceedings | 1998
Large-scale ab initio calculations based on three levels of parallelization
| British Library Online Contents | 2008