Research And Application Of Reverse Time Migration Based On CPU/GPU Heterogeneous Parallel Acceleration

With the deepening of oil and gas exploration,seismic migration imaging technology is one of the most direct and effective methods to obtain underground structure information;At present,seismic migration technology is mainly used in pre-stack seismic data processing.Prestack reverse-time migration is considered to be the most accurate imaging method to obtain underground structures,and it will produce a huge amount of computation and storage capacity in the stage of reverse-time migration processing.In addition,the oil and gas exploration is gradually developing from two-dimensional to three-dimensional exploration,which greatly increases the amount of seismic data calculation while enlarging the exploration scope and improving the exploration accuracy,and puts forward higher requirements for the calculation performance and efficiency of the cluster.In recent years,GPU acceleration technology has greatly improved the computational efficiency of seismic data processing.With the rapid development of large computing clusters,GPU clusters are mainly used in large-scale seismic data processing,but there are problems of heterogeneous nodes' collaborative computing.In this paper,prestack reverse time migration and high performance computing technology are combined to study how to improve the computing efficiency of reverse time migration data processing,focusing on parallel computing,task scheduling and storage optimization of CPU/GPU heterogeneous clusters.The main research results are as follows:1.Considering the heterogeneity of nodes in the cluster system and the parallelization characteristics of reverse time migration data processing,a framework model of reverse time migration data parallel processing is designed.The model adopts the adaptive node two-level task balancing scheduling mechanism,which realizes the reasonable allocation of first-level scheduling among nodes and the CPU/GPU second-level collaborative scheduling within nodes.2.In view of the problem of cluster computing resource utilization,this paper comprehensively considers the computing capacity and the assignment of computing tasks of nodes,studies the load balancing mechanism among nodes and the calculation method of node weight.A balanced scheduling algorithm for reverse time migration computing tasks in heterogeneous cluster environment is proposed,which completes the balanced scheduling of reverse time migration tasks among nodes and achieves a better load balancing effect3.Analyze the collaborative computing problem of heterogeneous cluster nodes,and study the parallel acceleration processing reverse time migration technology of CPU/GPU.The GPU acceleration technology is applied to reverse time migration algorithm to optimize the GPU storage strategy and reduce the data transmission between CPU and GPU.A method based on CPU multi-thread and multi-GPU card was proposed to accelerate the calculation of reverse time migration,which improved the calculation efficiency of reverse time migration data processing.4.Combined with the application of prestack reverse time migration,using the CPU/GPU heterogeneous parallel acceleration reverse time migration calculation method,the function of reverse time migration system is designed.Reverse time migration software was developed to realize the visual operation interface of prestack reverse time migration data processing to meet the practical business requirements.