Storage and performance issues in reverse time migration

Abstract

Prestack Reverse Time Migration (RTM) algorithm for depth imaging has now become feasible due to advances in high performance computing and clever programming techniques. RTM makes use of full acoustic two-way wave propagation algorithm for both forward and reverse time extrapolation. It is a computationally expensive process and the program run times are large in terms of CPU cycles and disk storage. The computational challenge is overcome by clever parallel programming techniques. As the forward propagated wavefield has to be stored at all time steps and read back for imaging, RTM needs a large amount of storage. This problem is solved by either using a large central storage or computing the forward wavefield twice. In this paper we propose the use of local disk space available on each node to write the forward wavefield and compare it with the other two methods. The use of local disk space reduces the communication time to the central storage and hence provides an improvement in performance. The performance is shown for 2D RTM algorithm.