Physical Data Design In Multiprocessor Database System

机器在过去的十年中，并行数据库系统由于其高性能、可伸缩性和可用性而越来越受欢迎。随着预测的未来数据库的大小和查询的复杂性，这些系统有效地利用数百或数千个处理器是必不可少的。一些研究已经反复证明，并行数据库系统的性能和可伸缩性取决于系统处理器间数据的物理布局。最近，提出了一种新的聚类策略——多属性网格聚类（MAGIC）。MAGIC优于当前的聚类策略，因为它可以使用两个或多个属性对关系进行分区。此外，它最大限度地减少了使用并行性的开销，从而允许系统扩展到大量处理器。该项目的重点是：(1)分析和扩展用于执行复杂查询的MAGIC，以及(2)在存在更新查询和不断变化的工作负载特征时合并用于动态在线重组关系的算法。该方法包括首先开发分析和仿真模型，然后是原型实现。本研究增强了并行数据库机的可扩展性，提高了其执行复杂查询的性能。

Machines During the past decade, parallel database systems havegained increased popularity due to their high performance,scalability, and availability characteristics. With thepredicted future database sizes and complexity of queries,it is essential that these systems effectively utilizehundreds or thousands of processors. Several studies haverepeatedly demonstrated that both the performance and scalability of a parallel database system is contingent on the physical layout of the data across the processors of the system.Recently, the design of a new declustering strategy, termed Multi-Attribute GrId deClustering (MAGIC), was introduced. MAGIC is superior to the current declustering strategies because it can partition a relation using two or more of its attributes. In addition, it minimizes the overhead of using parallelism to allow the system to scale to a large number of processors. This project focuses on: (1) analysis and extension of MAGIC for executing complex queries, and (2) incorporation of algorithms for dynamic on-line reorganization of a relation in the presence of update queries and changing workload characteristics.The approach involves development of analytical and simulation models first, followed by a prototype implementation. This research enhances the scalability of a parallel database machine and improve its performance for executing complex queries.