基于数据属性与服务的叠前深度偏移并行算法

Parallel algorithm for pre-stack depth migration based on data attributes and d ata serving

在三维叠前Kirchhoff深度偏移的大规模并行计算过程中,高效存储与获取射线旅行时数据是非常必要的。为此提出一种以"属性关联表"和"数据服务进程"为基础、基于分布存储的三维叠前Kirchhoff深度偏移并行算法。通过"属性关联表",在射线旅行时的计算阶段可大大提高存储效率;通过"数据服务进程",在地震道(集)的成像阶段可及时响应来自各个工作进程对所需射线旅行时数据的请求。实际测试结果表明,该并行算法具有较好的可扩展性,当在64个处理器上运行时并行效率仍高于84%。

Large-scale parallel comput at ion of 3-D pre-stack Kirchhoff depth migration requires efficient storage of t ravel-times. A parallel algorithm was developed for 3-D per-stack Krichhoff d epth migration based on distributed storage systems using an attribute relations hip table and data-serving procession method. The attribute relationship table method significantly improves the storage efficiency during computation of trave l-times. The data-serving processions respond quickly to requests for travel- times from slave processions during imaging of seismic records (groups). Test re sults with practical data verify the scalability of this parallel algorithm. Whe n this parallel algorithm is executed on a 64-cpu cluster, the efficiency is st ill above 84%.