Design and Evaluation of Algorithms for Storage Reclamation in Object Databases

对象数据库存储回收算法的设计与评估

• 批准号: 9521046
• 负责人: Alexander Wolf
• 金额: $ 34.44万
• 依托单位: University of Colorado at Boulder
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1995
• 资助国家: 美国
• 起止时间: 1995-09-01 至 1999-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9521046&HistoricalAwards=false
• 关键词: Design; Evaluation; Algorithms; Storage; Reclamation

A problem for an object database management system (ODBMS) is the reclamation of storage for persistent, secondary-memory objects that are no longer accessible. Although excellent reclamation algorithms have been developed for transient, primary-memory objects, these algorithms have been shown to be inappropriate for use on secondary-memory objects in object databases. This project investigates new algorithms for ODBMS storage reclamation. The critical aspects of secondary-memory reclamation are teased apart into a set of distinct policy areas. Rigorous trace-driven simulations of the algorithms are performed to gain an understanding of the relative performance of alternative policies within each area as well as an understanding of the interplay of policies across areas. Prototype implementations of the most promising algorithms are integrated into an existing ODBMS. This project makes three primary contributions to the ODBMS research, development, and user communities. First, a set of new reclamation algorithms appropriate for object databases is defined. Second, performance evaluations of the algorithms are used to reveal the circumstances under which the algorithms should be considered for adoption within a particular ODBMS or application. Third, a catalog of benchmark and live-application traces is created to establish an empirically founded characterization of application behavior that can be used far beyond evaluation of reclamation algorithms.

对象数据库管理系统（ODBMS）的一个问题是回收不再可访问的持久性二级内存对象的存储。虽然优秀的回收算法已经开发了短暂的，主要的内存对象，这些算法已被证明是不适合用于对象数据库中的辅助内存对象。本项目研究了ODBMS存储回收的新算法。二次记忆回收的关键方面被梳理成一组不同的政策领域。严格的跟踪驱动的模拟算法进行，以获得在每个区域内的替代政策的相对性能的理解，以及跨领域的政策的相互作用的理解。最有前途的算法的原型实现集成到现有的ODBMS。该项目对ODBMS的研究、开发和用户社区做出了三个主要贡献。首先定义了一组新的适用于对象数据库的回收算法。其次，性能评估的算法被用来揭示的情况下，该算法应被考虑采用在一个特定的ODBMS或应用程序。第三，创建基准测试和实时应用程序跟踪的目录，以建立应用程序行为的经验基础表征，其可用于远远超出回收算法的评估。