一种基于共享存储的叠前深度偏移并行算法

A Parallel Algorithm for Prestack Depth Migration Based on Shared Memory

为了解决叠前深度偏移计算量巨大这个问题,人们投入了很大精力来开发高效的并行算法.就此,在对一些相关并行算法进行分析之后,根据三维叠前Kirchhoff深度偏移的特点,提出了一种基于共享存储的简化并行算法.其中Slave进程直接存储和读取射线旅行时,消息传递总量显著减少;同时合理地结合“任务池”技术与粗粒度级并行,前者确保动态负载平衡,后者使得Slave与Master之间的通信开销尽可能小.实际教据的测试结果表明,此并行算法高效且可扩展性较好.

Prestack depth migration is computationally intensive. To deal with this problem, people have made big efforts to develop high efficient parallel algorithms. In this paper, after some parallel algorithms are analyzed, according to the character of 3-D Kirchhoff depth migration, a simplified algorithm is presented based on shared memory . In the proposed algorithm, the slave processes store and read ray traveltimes directly so that the total amount of message passing will be remarkably decreased. At the same time, this algorithm has well combined both the 'task pool' and the 'coarse granularity' techniques together, the former ensuring the dynamic load-balance and the latter reducing the cost of communication between the slaves and the master. In the end, experimental results show that this algorithm has high efficiency and scalability.