基于Spark的地震数据重建方法的并行化

Parallelization for Seismic Data Reconstruction Methods Based on Spark

地震勘探技术发展早已进入TB(terabytes)级数据时代,并逐步迈向PB(petabytes)级。为提升海量数据处理效率,将地震数据处理算法进行并行化是一种广泛采用的手段。但是一些复杂度较高的算法,诸如地震数据重建类方法等,并行化难度较大,加速效果不理想。Spark作为一种面向大数据处理的通用分布式并行计算技术,可以应用于并可简化地震数据处理算法并行化过程。借助于Spark的优势,通过两个实例讨论了基于Spark的地震数据重建并行化方法,提出了对于具有复杂输入输出组织数据方式的算法的并行化方法,提升了算法效率。研究成果为该类算法的Spark并行化开发提供了有益借鉴。

With the development of seismic exploration technology,more and more seismic data of terabytes or even petabytes are acquired and need to be processed.To improve the efficiency of massive data processing,parallelization of processing algorithms is a widely used solution.However,some algorithms with high complexity,such as data reconstruction methods,are so difficult to be parallelized that desired acceleration effect cannot be achieved.Fortunately,Spark,a general distributed parallel computing technology for big data processing,can make the parallelization of seismic data processing algorithms easier and more effective.By virtue of the advantages of Spark,the parallelization methods of seismic data reconstruction algorithms were discussed with two sample algorithms and approaches to deal with complicated organizations of input or output data were proposed.Efficiency of the algorithm were significantly improved by applying these methods.The research results provide useful references for the development of Spark parallelization for this kind of algorithms.