High-Performance Reverse Time Migration on GPU

Cabezas, Javier; Araya‐Polo, Mauricio; Gelado, Isaac; Navarro, Nacho; Morancho, Enric; María, José

doi:10.1109/sccc.2009.19

Cited by 14 publications

(16 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…However, this is only effective when the reading of the wavefields from disk is faster than the computations needed for one imaging time step. One way to achieve this is by compressing the wavefield before storing it to disk on a GPU or on a CPU (e.g., Cabezas et al, 2009).…”

Section: Migration In the Time Domainmentioning

confidence: 99%

See 1 more Smart Citation

Closing the performance gap between an iterative frequency-domain solver and an explicit time-domain scheme for 3D migration on parallel architectures

et al. 2014

View full text Add to dashboard Cite

Three-dimensional reverse-time migration with the constantdensity acoustic wave equation requires an efficient numerical scheme for the computation of wavefields. An explicit finitedifference scheme in the time domain is a common choice. However, it requires a significant amount of disk space for the imaging condition. The frequency-domain approach simplifies the correlation of the source and receiver wavefields, but requires the solution of a large sparse linear system of equations. For the latter, we use an iterative Krylov solver based on a shifted Laplace multigrid preconditioner with matrixdependent prolongation. The question is whether migration in the frequency domain can compete with a time-domain implementation when both are performed on a parallel architecture. Both methods are naturally parallel over shots, but the frequency-domain method is also parallel over frequencies. If we have a sufficiently large number of compute nodes, we can compute the result for each frequency in parallel and the required time is dominated by the number of iterations for the highest frequency. As a parallel architecture, we consider a commodity hardware cluster that consists of multicore central processing units (CPUs), each of them connected to two graphics processing units (GPUs). Here, GPUs are used as accelerators and not as an independent compute node. The parallel implementation of the 3D migration in frequency domain is compared to a time-domain implementation. We optimize the throughput of the latter with dynamic load balancing, asynchronous I/O, and compression of snapshots. Because the frequency-domain solver uses matrix-dependent prolongation, the coarse-grid operators require more storage than available on GPUs for problems of realistic size. Due to data transfer, there is no significant speedup using GPU-accelerators. Therefore, we consider an implementation on CPUs only. Nevertheless, with the parallelization over shots and frequencies, this approach could compete with the time-domain implementation on multiple GPUs.

show abstract

Section: Migration In the Time Domainmentioning

confidence: 99%

“…A second strategy is based on reducing the time needed to write the snapshots to disk, for instance, by asynchronous I/O (Cabezas et al, 2009) and wavefield compression. For the last, standard libraries with Fourier transformation or wavelet compression can be used (Ji et al, 2012).…”

Section: Introductionmentioning

confidence: 99%

Closing the performance gap between an iterative frequency-domain solver and an explicit time-domain scheme for 3D migration on parallel architectures

et al. 2014

View full text Add to dashboard Cite

show abstract

“…Este consiste en resolver una aproximación de la Ecuación (1) dos veces, estas dos soluciones reciben el nombre de propagaciones hacia adelante y hacia atrás, luego los dos resultados son correlacionados para obtener la imagen [4]. Matemáticamente este método implica el operador Laplaciano, diferencias finitas y un método explícito para realizar integración en el tiempo [14].…”

Section: Método Por Diferencias Finitasunclassified

“…Ahora bien, la pregunta para los investigadores en computación sísmica de alto rendimiento es: ¿cuál es el hardware más apropiado para implementar el proceso de Migración Sísmica (MS)? investigaciones recientes [3], [4], [5] muestran que tanto las GPGPU (General-Purpose Computing on Graphics Processing Units) como las en FPGAs (Field Programmable Gate Array) son dos opciones para acelerar el proceso de MS.…”

Section: Introductionunclassified

Reducción de los tiempos de cómputo de la Migración Sísmica usando FPGAs y GPGPUs: Un artículo de revisión

Fajardo

Villar

Pedraza³

2013

ing.cienc.

View full text Add to dashboard Cite

MSC:68M20 / PACS:93.85.Rt ResumenEste artículo hace una revisión entorno a los esfuerzos que actualmente se están realizando con el propósito de reducir el tiempo de cómputo de la MS. Nosotros introducimos los métodos más utilizados para realizar el proceso de Migración, así como también las dos arquitecturas computacionales que están ofreciendo mejores tiempos de procesamiento. Revisamos las implementaciones más representativas de este proceso sobre estas dos tecnologías y resumimos los aportes de cada una de estas investigaciones. El artículo finaliza con un análisis acerca de la dirección que deben tomar futuras investigaciones en esta área.Palabras clave: Evaluación del desempeño, FPGA, GPGPU, Métodos sís-micos de exploración, Migración Sísmica.

show abstract

“…These technologies are subject of major research in HPC since they have a better perspective of computing [10,14]. Different implementations of SM algorithms have been developed using those alternatives platforms [4,12,17,21,22,29]. Results show a reduction in the running times of SM algorithms, leading to combine these alternative platforms with traditional CPU clusters in order to get a promising future in HPC for seismic exploration.…”

Section: Introductionmentioning

confidence: 99%