并行数据库中数据分布和查询处理技术的分析-analysis of data distribution and query processing technology in parallel database.docxVIP

摘要并行数据库系统作为未来高性能的数据库管理系统，已经成为数据库领域 中重要的研究课题。并行数据库系统的性能与数据库中的关系在各个节点的分 布密切相关，本文针对一维数据分布方法不适于复杂查询要求，而多数多维数 据分布方法难于实现的缺点，分析并比较了各种基于树结构的多维数据分布方 法，给出了基于k．d树结构的分布方法的详细设计方案，指出设计基于传统索 引结构的数据分布方法是解决并行数据库系统数据分布问题的一种较好方式。 该方法在构造树的过程中，完成了对数据空间的划分和数据超方体的放置，不 仅可以使各个处理机能够合理得分配工作负载，而且可以方便地利用索引执行 查询处理操作，提高了系统的效率。如何将并行数据库系统应用于局域网络环境，并实现查询处理及优化也是 本文研究的工作之一。本文采用了基于工作站网络的数据库的并行查询处理设 计思想，在传统的串行数据库基础之上加入了并行化的设计方案，通过开发关 系数据库的三个固有并行性来提高系统执行查询处理的能力。在此基础上，本 文提出了一种并行查询处理的结构，并根据该结构在局域网络环境下设计和实 现了并行查询处理的模拟系统。通过对实验结果的分析与比较，证明了合理的 并行化设计可以减少通信带来的开销，获得较好的性能。关键字：多维数据分布树结构查询处理工作站网络AbstractAs high performance database systems in the future，the study of parallel database systems has been a very important research problem in the field of database． Data declustering methods that place relation fragments at different network sites have significant effect on the performance of parallel databases．But single dimensional declustering methods are not suitable to the requests of complicated query,and most multidimensional declustering methods are too complicated to implement．So，this paper will research several multidimensional data declustering methods based on the tree structures，and describe an algorithms based on the k-d tree structure．It is pointed out that the methods based on the traditional index structure are very important ways to solve the data storage problem in the parallel database systems．During the process of constructing the tree，this method has achieved the goal of partitioning of the data space and placing of the data hyper-cubes，which not only balances the load of every processor,but also executes query operation by index structure tO get higher system efficiency．The other focus of this paper is on the way to apply parallel database systems to LAN and to implement query processing and optimizing．In this paper,we utilize the idea of using networks of workstations for parallel query processin，．gof applyingparalleltechniquesonthetraditionalseriesdatabasesWeexploit the three parallelism of relatio