Analysis of Parallel and Vectorized Algorithms of Pre-Stack Kirchhoff Time Migration in Virtual Environments

• Rodrigo Alves Prado da Silva UFF
• Cristiana Barbosa Bentes UERJ
• Lúcia Maria de Assumpção Drummond UFF

Resumo

One of the most popular methods of seismic migration is the Pre-Stack Kirchhoff Time Migration (PKTM) which is computationally intensive. Because cloud computing has proven to be a promising alternative for running high-performance computing applications, the impacts introduced by virtualization layers, which typically support cloud services, on the different implementations of the PKTM are evaluated in this paper. Thus, the performance of different parallel versions of PKTM, using MPI and OpenMP, and vectorized, automatically or manually when performed with virtualization technologies commonly used in clouds: KVM, Docker and Singularity were analyzed. Our results showed that, in relation to all PTKM versions, vectorized or not, all virtualization technologies tested introduced small overheads, with differences in execution times of approximately 2% on average when compared to running without virtualization. It was also possible to observe that virtualization based on Singularity and Docker performed better than the one managed by KVM.

Palavras-chave: Virtualization, Cloud computing, Computer science, High performance computing, Systems engineering and theory, Parallel processing, Kirchhoff, Parallelism, Vectorizing