Parallel processing of Prestack Kirchhoff Time Migration on a PC Cluster

This paper discusses an approach that implements a parallel processing of 3-D Prestack Kirchhoff Time Migration (PKTM) on a low-cost PC Cluster by using the Message Passing Interface (MPI), and analyses its performance using a real seismic data as examples. The PC Cluster provides a significant acceleration of the migration processing with the exact same image quality. The ratio between the communication time and processing time is a critical indicator for determining the efficiency of the PC Cluster. If the processing time is longer than the communication time, using more CPUs can efficiently reduce the elapsed time. On the contrary, using more CPUs cannot reduce the elapsed time. Appling this approach to the Alba dataset on our PC Cluster up to 15 CPUs, the elapsed time of PKTM is inversely proportional to the number of CPUs used. The elapsed time for migrating a 2-D seismic line is reduced from 15 h using one CPU to 1 h using 15 CPUs. The elapsed time for migrating a 3-D image is reduced from 630 h using one CPU to 42 h using 15 CPUs. Further reduction can be achieved by using more CPUs. However, an optimal CPU number is expected for an application on large PC clusters with hundreds of nodes. Adapting existing algorithms to the cluster environment offers the potential to allow the application of more accurate algorithms for PKTM to construct a more accurate image. This work has proven that the PC Cluster is a powerful and scalable computing resource for oil and gas exploration organizations.