摘要
为了适应高性能硬件资源对地震成像技术的挑战,并满足百TB级宽方位高密度地震数据的深度域叠前成像需求,本文深刻分析了积分法深度域叠前成像技术的技术瓶颈,突破了多维多任务拆分、动态异步任务调度、数据重复访问导致的I/O瓶颈、CPU/GPU异构硬件协同计算等瓶颈技术。提出了一套高效的叠前成像方法和策略,研发了混合域并行、千节点异构并行框架、CPU+GPU高性能协同计算、动态任务调度、高效高压缩比旅行时表压缩、OVT域叠前成像和Q叠前成像等技术,形成了GeoEast积分法叠前深度偏移软件。该软件具备起伏地表、TTI各向异性、OVT分方位偏移、Q偏移、OBN数据的双基准面偏移和镜像偏移等功能;可输出炮检距道集、反射角度道集或构造倾角道集;具备CPU/GPU大规模并行和断点保护、异常节点自动处理等能力,计算效率业界领先,在256及以上节点并行,其加速比接近线性。
To meet the challenge of high performance hardware resources to seismic imaging technology,and to meet the needs of the depth domain prestack imaging of 100 TB wide azimuthorhigh density seismic data,The paper analysis the technical bottleneck of depth domain prestack imaging technology deeply.Breaking through the bottleneck of multitask splitting,dynamic asynchronous task scheduling,I/O bottleneck,CPU/GPU heterogeneous hardware co calculation,and so on.proposed a set of effective prestack imaging methods and strategies.Developed a hybrid domain parallel,1000 nodes heterogeneous parallel framework,CPU+GPU high performance collaborative computing,dynamic task scheduling,high efficiency and high compression ratio of travel time table compression,OVT domain prestack imaging,Q prestack imaging and other techniques,formed GeoEastkirchhoff PSDM software.The software has the undulate surface migration,TTI anisotropy migration,OVT domain migration,Q migration,double reference surface migration for OBN data and mirror migration function;output offset domain gather,reflection angle domain gather or structure angle domain gathers.it also has CPU/GPU large scale parallel,breakpoint protection,automatic processing of abnormal nodes,etc.The KPSDM implementation can obtain close to linear speedup when it processes real word data on a 256 or above nodes cluster.
引文
[1]Changhai Zhao,Jianlei Zhang,.etc,Hybrid domain parallel algorithm for 3dKirchhoff Prestack depth migration,AAPG,1140-115,2016.
[2]赵长海、张建磊、罗国安,3D Kirchhoff叠前深度偏移的混合域并行算法,HPC,624-627,2015.
[3]张建磊、王成祥、王狮虎,高效积分法叠前深度偏移,石油地球物理勘探,45(Z1),58-61,2010.
[4]F.Cappello,A.Geist,W.Gropp,S.Kale,B.Kramer,and M.Snir,"Toward exascale resilience:2014update,"Supercomputing frontiers and innovations,vol.1,2014.
[5]J.Dean and S.Ghemawat,MapReduce:simplified data processing on large clusters,Commun.ACM,vol.51,pp.107-113,2008.
[6]C.Qi,L.Cheng,and X.Zhen,Improving MapReduce Performance Using Smart Speculative Execution Strategy,Computers,IEEE Transactions on,vol.63,pp.954-967,2014.