On improving performance of migration algorithms using MPI and MPI-IO
Dheeraj Bhardwaj, Suhas M. Phadke, Sudhakar Yerneni · 2000
High level of I/O performance is necessary in making use of parallel machines for many scientific applications. In this paper we discuss the I/O requirements of one such application: 3D seismic migration. I/O performance is the bottleneck rather than the computational or communication performance in 3D seismic imaging. For large 3D data volume it is not possible to read and keep all the required data and information in computer memory. Therefore the data are partially read and intermediate results are written out at various stages during the execution of the code. In this article we have discussed an approach to handle the massive I/O requirements of seismic migration using MPI-IO. We have also optimised some data communication using MPI calls. The performance of parallel seismic migration code on PARAM 10000, which is a cluster of SUN workstations, is discussed for two data sets.