Pre-stack kirchhoff time migration on hadoop and spark

Abstract

Pre-stack Kirchhoff time migration (PKTM) is one of the most widely used migration algorithms in seismic imaging area. How- ever, PKTM takes considerable time due to its high computational cost, which greatly affects the working efficiency of oil industry. Due to its high fault tolerance and scalability, Hadoop has become the most popular platform for big data processing. To overcome the shortcoming too much network traffic and disk I/O in Hadoop, there shows up a new distributed framework—Spark. However the behaviour and performance of those two systems when applied to high performance computing are still under investigation. In this paper, we proposed two parallel algorithms of the plrestack Kirchhoff time migration based on Hadoop and Sark respectively. Experiments are carried out to compare the performances of them. The results show that both of implementations are efficient and scalable and our PKTM on Spark exhibits better performance than the one on Hadoop.