并行算法

In this paper, an extended divide and conquer algorithm is intended proposed, which is for solving the real symmetric band generalized eigenvalue problem under distributing environment Eigenvalue partition theorem is presented and proved Based on divide and conquer by extension, this algorithm computes generalized eigenpaires of symmetric band matrix pencil by bisection and generalized Rayleigh quotient iteration Theoretic analysis and numerical results show that this algorithm is better than the classic software package LAPACK when bandwidth is small and the scale is large Combined with multisection, which has good parallelism, it got good effects under distributed environments

提出了分布式环境下计算对称带状广义特征值问题的一种扩展分治算法，给出了特征值分割定理及其证明算法在扩展分治的基础上，利用二分压缩结合广义Rayleigh商迭代计算广义特征对理论分析和数值实验表明，对于窄带宽大规模的广义特征值问题，该分治算法明显优于LAPACK软件包结合并行性好的多分法，在分布式环境下获得了很好的并行效果1 引言本文研究了对称带状广义特征值问题Ax =λBx ( 1)的并行计算，其中，A ，B均为半带宽为r的n阶实对称带状矩阵且其中之一是正定的本文总假设B是正定的求解此问题有两种传统方法，第1种方法是通过计算矩阵B的Cholesky分解，将问题( 1)转化为标准特征值问题[1～3] ，进一步

The traditional query tree optimization methods,parallel database optimization methods based on left linear trees and right linear trees,bushy trees,and operation of the forest,have their own pros and cons,they have been more in-depth and maturity of the study.The query optimization method based on multiple weighted tree has studied its model of parallel query plan,its complexity model of parallel query plan and query optimization algorithms.The semantic query method transforms an inquiry into one or several semantic equivalence inquiries then has to find and implement a strategy to achieve a better query.Agent-based parallel database query optimization using Multi-Agent technology to automatically search the integrity constraint conditions which are related to the determined query,there for,the efficiency between several relations' joins has been greatly improved.The parallel optimization algorithm,based on genetic algorithm which is suitable for multi-joins of cluster environment,has deeply studied the relations storage options,multi-joins query optimization and query processing and other key technologies based on cluster parallel database.

传统的查询树优化方法，即基于左线性树、右线性树、浓密树、操作森林的并行数据库查询优化方法，各有优劣，对其的研究比较深入、成熟；基于多重加权树的查询优化方法，研究了其并行查询计划模型、并行查询计划的复杂性模型和查询优化算法；语义查询优化方法将一个查询变换成一个或数个语义等价的查询，进而寻找并执行这些等价查询中具有较好实现策略的一个；基于Agent的并行数据库查询优化采用Multi-Agent技术自动查找与给定查询有关的完整性约束条件，使得多个关系间连接操作的效率得到很大的提高；基于遗传算法的并行优化算法，深入研究了基于机群并行数据库中关系存储的选择、多连接查询优化和查询处理等关键技术。

Design Krylov subspace conjugate residual squared algorithm for nonsym-metric linear systems and the improved conjugate residual squared algorithm for distributed parallel computing based on biconjugate residual method algorithm, and give theoretical analysis and comparisons of algorithms for two algorithms.

基于BiCR算法设计了求解非对称线性方程组Krylov子空间平方共轭残差算法和适合分布式并行计算的改进的平方共轭残差算法，并对两种算法进行了理论分析和算法比较，串行和并行数值试验表明所提方法具有较好的收敛速度和并行性能。

Aiming at solving these problems, this paper proposed a parallel algorithm with multi-transmitting redistributed data.

针对这些问题，提出了一种多次传送重新分配数据的并行算法。

The main academic productions of this thesis are:(1) the parallel algorithms of computing permanents and permanental polynomials of sparse matrices are designed and implemented;(2) an efficient balancing straggles of parallel load is proposed using the theory of scheduling;(3) valuable data for the research of the molecule structure of fullerenes are obtained;(4) the multi-dimensional stable marriage model and its numerical methods give a new path to solve the clustering problems.

本文的主要学术贡献是：(1)设计并实现了稀疏矩阵积和式与积和多项式的并行算法；(2)利用排序理论给出了有效的负载平衡策略；(3)为富勒烯分子结构的研究提供了有价值的数据支持；(4)多维稳定婚姻模型及计算方法，为聚类问题的求解提供了新途径。

ABSTRACT The dissertation concentrates on the design and implementation of a high speed modular graphics system based on Transputer includi...

论文的主要工作是设计和实现了基于Transputer的模块化的高速图形系统及其并行算法的研究。

ABSTRACT The dissertation concentrates on the design and implementation of a high speed modular graphics system based on Transputer including research in parallel graphics processing algorithms.

论文的主要工作是设计和实现了基于Transputer的模块化的高速图形系统及其并行算法的研究。

RapidIO or Ethernet is adopted as the system interconnection network with star-connected topology.

并根据模型预测控制的计算特点，研究并提出了适用于该计算的基于按行—维划分输出矩阵的并行算法。

In the end, experimental results show that this algorithm has high efficiency and scalability.

实际数据的测试结果表明，此并行算法高效且可扩展性较好。

In the third parts, by using high accurate methods and HPF parallel computing, the N-S equations under cylindrical coordinates are used to simulate space developing compressible axisymmetric jet.

第三部分采用高精度的差分格式求解柱坐标下的三维N—s方程，利用HPF并行算法对空间发展的可压缩轴对称射流进行直接数值模拟。

The split between the two groups can hardly be papered over.

这两个团体间的分歧难以掩饰。

This approach not only encourages a greater number of responses, but minimizes the likelihood of stale groupthink.

这种做法不仅鼓励了更多的反应，而且减少跟风的可能性。