Enhancing Reverse Time Migration: Hybrid Parallelism plus Data Compression

Resumo

Migration techniques like the Reverse Time Migration (RTM) are time-consuming and data-intensive.
RTM is time-consuming due to the computational cost associated with the stability and dispersion conditions in
the discrete two-way wave equation. On the other hand, RTM is data-intensive because of the number of input
seismograms to be loaded and the need to store temporary files, such as the forward propagated wavefields, to build
the migrated seismic images and partial migrated images before stacking. Thus, to overcome the time-consuming
problem, we implement a Reverse Time Migration algorithm that explores the Message Passing Interface (MPI)
library and the OpenMP+vectorization to explore parallelism in hybrid architectures. On the other hand, we
implement compression techniques to reduce storage and to improve network data transfer. The data compression
can be used to reduce the storage of seismograms, model parameters, migrated seismic images, and temporary files.
Results show that an MPI/OpenMP+vectorization strategy is efficient in hybrid multi-core machines. Besides, high
levels of data compression suggest that its use does not hamper the final migrated seismic images.