Solution of A P and S wave propagation model using high performance computation

The propagation of seismic waves is affected by the type of transmission media. Therefore, it is necessary to solve a differential equation system in partial derivatives allowing for identification of waves propagating into an elastic media. This paper summarizes a research using a partial different...

Full description

Autores:
Tipo de recurso:
Fecha de publicación:
2019
Institución:
Universidad de Medellín
Repositorio:
Repositorio UDEM
Idioma:
eng
OAI Identifier:
oai:repository.udem.edu.co:11407/5690
Acceso en línea:
http://hdl.handle.net/11407/5690
Palabra clave:
Asynchronous copies and executions
Elastic media
GPU constant memory
GPU shared memory
Modelling
PML
Graphics processing unit
Models
Shear waves
Wave propagation
Asynchronous copies and executions
Computational architecture
Constant memory
Differential equation systems
Elastic media
Finite differences methods
High performance computation
Shared memory
Memory architecture
Rights
License
http://purl.org/coar/access_right/c_16ec
Description
Summary:The propagation of seismic waves is affected by the type of transmission media. Therefore, it is necessary to solve a differential equation system in partial derivatives allowing for identification of waves propagating into an elastic media. This paper summarizes a research using a partial differential equation system representing the wave equation using the finite differences method to obtain the elastic media response, using an staggered grid. To prevent reflections in the computational regions, absorbent boundaries were used with the PML method. The implementation of the numerical scheme was made on two computational architectures (CPU and GPU) that share the same type of memory distribution. Finally, different code versions were created to take advantage of the architecture in the GPU memory, performing a detailed analysis of variables such as usage of bandwidth of the GPU internal memory, added to a version that is not limited by the internal memory in the graphic processing unit, but rather by the memory of the whole computational system. © 2019 Ecopetrol S.A.. All rights reserved.