Multi-GPU 3-D Reverse Time Migration with Minimum I/O.

CARLA(2022)

引用 0|浏览2
暂无评分
摘要
Seismic imaging techniques based on two-way wave equations are computationally and data-intensive activities in the oil and gas industry. For instance, Reverse Time Migration (RTM) migrates a set of SEG-Y format data from the disk called a seismogram. Besides, during execution, the RTM application needs to store the forward-propagated wavefield (or source wavefield) on disk to build the final seismic image. Storing the source wavefield for multiple RTMs executing in parallel is even more challenging because the storage capacity can reach tens of Terabytes of information. Aiming to mitigate the storage demand, we develop a 3-D RTM with source wavefield reconstruction. The source wavefield is reconstructed by introducing a new wave equation to the problem and adjusting the initial and boundary conditions to take advantage of random boundary conditions' (RBC) properties. The RBC does not suppress unwanted waves coming from the artificial boundary enabling full wavefield recovery. We also develop a hybrid OpenACC/MPI implementation for the 3-D RTM on a multi-GPU machine. We test the RTM implementation on the 3-D HPC4E Seismic Test Suite. The numerical experiments show that the OpenACC/MPI 3-D RTM, which implements the wavefield reconstruction, presents the best execution times and hard disk demands.
更多
查看译文
关键词
High performance computing, Reverse time migration, Wavefield reconstruction, OpenACC/MPI implementation
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要