Database Processing on a Cube-Connected Multicomputer System(Computer and Information Science)

本项目研究多计算机系统中数据库处理的新算法。用于数据库处理的并行架构需要同时包含并行处理器（CPU）和并行I/O（磁盘访问）功能。本研究的重点是布尔n-立方体连接架构，每个节点连接一个磁盘，从而提供并行I/O。支持并行I/O的需求产生了两个重要问题：数据组合和非统一数据分布。由于多维数据集互连包含许多其他结构，如树、环等，因此可以利用此属性有效地支持各种数据库操作中的数据组合，如选择、聚合、连接和投影。本研究使用一个64节点的超立方体，对各种初始和中间数据分布条件下的各种数据库操作进行了分析和模拟研究。现代制造技术使得以低成本生产数百甚至数千个计算元件（处理器）成为可能。这项研究的意义在于，并行处理算法对于利用制造潜力为数据库处理等应用产生真正的好处至关重要。在新研究人员研究启动活动项下提供支助。

This project investigates novel algorithms for database processing in multicomputer systems. Parallel architectures for database processing need to incorporate both parallel processors (CPU's) and parallel I/O (disk access) capability. This research is focused on a boolean n-cube connected architecture, with a disk attached to each node, thereby providing parallel I/O. The need to support parallel I/O gives rise to two important issues: data combination and non-uniform data distribution. Since the cube interconnection subsumes many other structures such as trees, rings, etc., this property is exploited to efficiently support data combination in a variety of database operations such as Selection, Aggregation, Join, and Projection. The research involves both analytic and simulation study of various database operations under a variety of initial and intermediate data distribution conditions, using a 64-node hypercucbe. Modern fabrication techniques make it possible to produce hundreds or even thousands of computing elements (processors) at low cost. The significance of this research is that parallel processing algorithms are critical to harnessing the fabrication potential to produce real benefit for applications such as database processing. Support is provided under the Research Initiation activity for New Investigators.