Parallel computation in reflection seismology

Abstract

Since the introduction of digital data processing in reflection seismology, new algorithms have been developed, in pace with the progress of computer power. Massively parallel computers have appeared in the oil industry, and it has been recognized that parallel computation is very attractive in cost performance. In this paper we investigate algorithms in parallel computation for seismic modeling and pre-stack depth migration. The pseudospectral method which we adopted for the seismic modeling requires a 2-D FFT. Each processor can contain only part of the data, therefore a 2-D FFT cannot be achieved in one pass. Since data swapping is necessary, we developed a parallel swapping algorithm and a parallel disk I/O method to reduce the computation time. There are many theoretical algorithms to apply pre-stack depth migration. For parallel computation we adopted the explicit extrapolation finite difference method in the frequency and space domain. Implementation was done using “parallel frequency slice” computations. All developed algorithms are easy to implement in workstation clusters since these algorithms are based on the message passing method.