Abstract: Caching is one of the most fundamental topics in modern computing. There exist many buffer replacement strategies， which are widely used in operating systems， file systems， databases， storage systems， etc. A performance evaluation methodology to evaluate buffer replacement strategies for RDBMS is proposed. This methodology adopts direct measurement way and employs five performance metrics including the ones newly proposed by this paper and those seldom considered in other publications due to their simulation method. Significance tests and a joint way to use these tests for cross-strategy comparison are presented and various workloads are designed and used in order to effectively evaluate buffer replacement strategies in both single-user and multiuser environments. Using this methodology， it performs a detailed and comprehensive experimental study about the performance of thirteen representative buffer replacement strategies in a real RDBMS environment. From the experimental results， some useful conclusions are made.

Key words: buffer replacement strategy, performance evaluation methodology, significance test, access pattern, TPC-H and TPC-C benchmark

摘要： 缓冲区是计算机领域一个非常重要的研究主题。现在存在很多缓冲区置换策略，广泛地用于操作系统、文件系统、数据库以及存储系统中。提出了一种性能评测方法来评测关系数据库的缓冲区置换策略。这种方法采用直接测试方式，使用五个性能指标，包括新提出的指标以及由于模拟方法的局限性在以前的研究中很少使用的性能指标，同时利用显著性测试并提出一种联合使用多个显著性测试的方法进行置换策略之间的比较，设计和使用多种工作负载在单用户和多用户情况下有效地评测各种策略。使用这种评测方法，在一个真实的关系数据库环境中，对十三种典型的缓冲区置换策略的性能进行了详细、全面的实验研究，从中可以获得一些有用的结论。

关键词: 缓冲区置换策略, 性能评测方法, 显著性测试, 访问模式, TPC-H和TPC-C基准