High‐performance computing of 3D blasting wave propagation in underground rock cavern by using 4D‐LSM on TianHe‐3 prototype E class supercomputer
Abstract Parallel computing assigns the computing model to different processors on different devices and implements it simultaneously. Accordingly, it has broad applications in the numerical simulation of geotechnical engineering and underground engineering, of which models are always large‐scale. W...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2022-09-01
|
| Series: | Deep Underground Science and Engineering |
| Subjects: | |
| Online Access: | https://doi.org/10.1002/dug2.12015 |
| Summary: | Abstract Parallel computing assigns the computing model to different processors on different devices and implements it simultaneously. Accordingly, it has broad applications in the numerical simulation of geotechnical engineering and underground engineering, of which models are always large‐scale. With parallel computing, the computing time or the memory requirements will be reduced by splitting the original domain of the numerical model into many subdomains, which is thus named as the domain decomposition method. In this study, a cubic and equal volume domain decomposition strategy was utilized to realize the parallel computing on the distributed memory system of four‐dimensional lattice spring model (4D‐LSM) based on the message passing interface. With a more efficient communication strategy introduced, this study aimed at operating an one‐billion‐particle model on a supercomputer platform. The preprocessing procedure of the parallelized 4D‐LSM was restructured and the particle generation strategy suitable for the supercomputer platform was employed to minimize the time consumption in preprocessing and calculation. On this basis, numerical calculations were performed on TianHe‐3 prototype E class supercomputer at the National Supercomputer Center in Tianjin. Two field‐scale three‐dimensional blasting wave propagation models were carried out, of which the numerical results verify the computing power and the advantage of the parallelized 4D‐LSM in the simulation of large‐scale three‐dimension models. Subsequently, the time complexity and spatial complexity of 4D‐LSM and other particle discrete element methods were analyzed. |
|---|---|
| ISSN: | 2097-0668 2770-1328 |