Parallel Implementation of Pre-stack Least-squares Migration for Distributed Systems

Summary Least-squares wave-equation pre-stack depth migration is a computationally expensive algorithm, requiring repeated application of wave-field modeling and migration algorithms. We describe a parallel implementation of least-squares migration for distributed parallel computing systems using MPI. The implementation distributes both the storage and processing requirements. In storage, the migrated data is distributed across the memory spaces of the MPI processes using the depth dimension. Likewise, the data (before migration) is distributed across the memory spaces of the MPI processes using the frequency dimension. In processing, the computational work is shared across MPI processes by partitioning the frequency dimension. We apply the algorithm to the Marmousi finite difference data-set, and find that the computation time of the algorithm scales with the number of MPI processes.