机群计算在基于有限元法电法勘探中的应用

【作者】 章文；

【导师】 李代平；

【作者基本信息】 广东工业大学 ， 计算机应用技术， 2003， 硕士

【摘要】 高速计算机网络和不断提高的微处理芯片性能使得计算机网络成为吸引人的并行计算载体。仅依赖于商业化的硬件和软件，计算机网络能够提供高性价比、高可用性的计算。这种高性能计算潮流一般被称为机群计算。近年来，机群计算已经成为学术界和业界团体研究的热点。 并行虚拟机(PVM)是一个可以将异构的“机群计算机”集合虚拟成一台高性能并行机的软件包。这样大规模计算问题可以通过使用众多计算机聚集取来的计算能力和内存来低成本解决。PVM可移植性好，可运行在当今的多数平台上。PVM拥有着大量的用户，现已成为世界上分布式计算事实上的标准。 本文成功的将机群计算应用到解决在电法勘探中使用有限元方法(FEM)时产生的大规模线性方程组问题。有限元被证明为在许多科学和工程领域中连续物理系统模拟的一种普遍方法，但缺点是传统的串行计算机需要花费太多的计算时间。FEM需要对域进行离散化(一个网格或者模型)，FEM的这种思想和机群分而治之的思想相似。因此，有限元节点代码可以被并行化、在不同的基于PVM机群结点上运行。程序用C++编写，通过调用PVM库函数提供对并行虚拟机的访问。有限元法产生的线性方程组使用共轭梯度算法求解。共轭梯度(CG)法与有限元法联系的十分自然因为两者都是采用相似的最小化过程。共轭梯度算法不需要构造全局矩阵，它在大规模并行结构中能被有效地实现。更多还原

【Abstract】 High speed network and improved microprocessor performance are making networks of computers an appealing vehicle for parallel computing. By relying solely on commodity hardware and software, network of computers can offer cost-effective high performance and high availability computing. This new wave in high performance computing is popularly called as Cluster Computing. In recent years, Cluster computing has become a hot topic research among academic and industry community.Parallel Virtual Machine (PVM) is a software package that permits a heterogeneous collection of "cluster computers’ to function as a single high-performance parallel machine. Thus large computational problems can be solved more cost effectively by using the aggregate power and memory of many computers. PVM is portable and runs on a wide variety of modern platforms. With tens of thousands of users, PVM has become the de facto standard for distributed computing world-wide.This dissertation successfully uses cluster computing for solving large-scale linear algebraic equations problems when the finite element method (FEM) is applied in electrical prospecting. FEM has proven to be a versatile method for the simulation of continuous physical systems in many areas of science and engineering. But a disadvantage arises due to the very high computation times regarding conventional serial computers. FEM requires a discretization (a mesh or a model) of a domain. This thought of FEM is similar to that of cluster computing, which is divide and rule. Therefore, the finite element code can be parallelized and executed on different cluster nodes on which PVM is available. The program is written in C++, provided access to PVM through the use of calls to PVM library routines. The system of linear equations produced by the FEM is solved by conjugate gradient (CG) algorithm, which is associated naturally with finite element problems because both finite element and conjugate gradient use similar minimization procedures. The CG algorithm does not require construction of the global matrix. It can be implemented efficiently on aAbstractmassively parallel architecture.更多还原