Principles and Implementation Techniques for the Recovery Component of a DBMS (Computer and Information Science)

这项研究调查了交易系统在具有高度并行性的计算机系统上的性能极限。数据库管理系统的性能由其恢复组件决定。拟议的工作将涉及最大限度地减少数据库管理系统恢复部分所需时间的方法。要调查的方法有两种：一种是使用保证可恢复性的“非阻塞”方法。另一种类型使用事务语义的知识，包括针对不同抽象级别分别实现并发控制和恢复。这项工作的重要性在于，银行、金融市场和预订系统等许多行业都在大量使用大规模共享数据库。实现这一点的数据库必须快速、正确地更新，即使许多客户正在访问相同的数据。此外，即使某些计算机组件出现故障，共享数据库也必须可靠地运行。在这里进行的研究将导致减少可靠地处理请求所需的时间，低于目前最先进的系统所需的时间。

This research investigates the limits of transaction system performance on computer systems with a high degree of parallelism. The performance of a database management system is dominated by its recovery component. The proposed work will address methods of minimizing the time taken by the recovery component of a database management system. The methods to be investigated are of two types: one type is the use of "non-blocking" methods of guaranteeing recoverability. The other type uses knowledge of transaction semantics, including the implementation of concurrency control and recovery separately for different levels of abstraction. The importance of this work is that many industries such as banks, financial markets, and reservation systems are making heavy use of large-scale shared databases. The databases that make this possible must be updated quickly and correctly, even when many customers are accessing the same data. Also, the shared databases must function reliably, even when some of the computer components fail. The research conducted here will lead to reducing the time required to reliably process a request, below the time required by present day state-of-the-art systems.