Design and Implementation of a Shared Memory Multiprocessor Database Machine

共享内存多处理器数据库机的设计与实现

• 批准号: 8715235
• 负责人: Michael Stonebraker
• 金额: $ 134.76万
• 依托单位: University of California-Berkeley
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1988
• 资助国家: 美国
• 起止时间: 1988-03-01 至 1991-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=8715235&HistoricalAwards=false
• 关键词: Design; Implementation; Shared; Memory; Multiprocessor

The goal of this project is to research and build a parallel input/output (I/O) system for a shared-memory multiprocessor. The main applications intended for this system are parallel database operations with a target performance of 1000 transactions per second. This project is a natural extension of work done at the University of California, Berkeley in shared-memory multiprocessor architecture (SPUR project) and database management systems (INGRES and POSTGRES). While many research results have been obtained in the area of processor and memory design by a number of research groups, considerable less effort has been directed towards the issues of I/O design. As processor speeds increase, it becomes more obvious that I/O operations constitute the bottleneck of parallel systems. Disk technology is now at the stage where it can provide cost effective high I/O bandwidth. The specific research objectives are: (1) to investigate new ways to couple large numbers of I/O devices to multiprocessors without affecting the limited memory system bandwidth; (2) to explore architectures for I/O processors; and (3) to implement hardware assists for database applications such as stable memories, string manipulation co-processors and hardware coherency and logging mechanisms. This proposal is strongly recommended for funding because: (1) the principal investigators research records are excellent and the research environment is extremely stimulating--thus there is a high chance of success; (2) there is a great need for research in the parallel processing area covered by this project and it is expected that the impact of this research will be significant.

这个项目的目标是研究和构建一个用于共享内存多处理器的并行输入/输出(I/O)系统。该系统的主要应用程序是并行数据库操作，目标性能为每秒1000个事务。该项目是加州大学伯克利分校在共享内存多处理器体系结构(SPUR项目)和数据库管理系统(Ingres和Postgres)方面工作的自然延伸。虽然许多研究小组在处理器和存储器设计领域已经取得了许多研究成果，但针对I/O设计问题的努力却相当少。随着处理器速度的提高，I/O操作构成并行系统的瓶颈变得越来越明显。磁盘技术现在处于可以提供经济高效的高I/O带宽的阶段。具体的研究目标是：(1)研究在不影响有限的存储系统带宽的情况下将大量I/O设备耦合到多处理器的新方法；(2)探索I/O处理器的体系结构；(3)为数据库应用实现硬件辅助，如稳定存储器、串操作协处理器以及硬件一致性和日志机制。强烈建议为这项建议提供资金，因为：(1)主要调查人员的研究记录极佳，研究环境极具启发性--因此成功的机会很高；(2)在该项目所涵盖的并行处理领域非常需要研究，预计这项研究的影响将是巨大的。