数据库系统参数调优方法综述  

Survey of Approaches to Parameter Tuning for Database Systems

作　　者：曹蓉 鲍亮 崔江涛[1] 李辉[1] 周恒 Cao Rong;Bao Liang;Cui Jiangtao;Li Hui;Zhou Heng(School of Computer Science and Technology,Xidian University,Xi’an 710071;Inspur Group Co.,Ltd.,Jinan 250014)

机构地区：[1]西安电子科技大学计算机科学与技术学院,西安710071 [2]浪潮集团有限公司,济南250014

出　　处：《计算机研究与发展》2023年第3期635-653,共19页Journal of Computer Research and Development

基　　金：国家自然科学基金项目(62172316);教育部人文社会科学研究项目(17YJA790047);陕西省软科学研究计划项目(2020KRZ018);陕西省哲学社会科学重大理论与现实问题研究项目(20JZ-25);陕西省重点研发计划项目(2019ZDLGY13-03-02);河北省重点研发计划项目(20310102D);陕西省自然科学基金项目(2019JM-368)。

摘　　要：数据库系统具有大量的配置参数,参数配置不同会导致系统运行时很大的性能差异.参数优化技术通过选择合适的参数配置,能够提升数据库对当前场景的适应性,因此得到国内外研究人员的广泛关注.通过对现有的数据库参数调优方法进行总结分析,根据参数优化方法是否具有应对环境变化的能力,将现有工作分为固定环境下的数据库参数优化方法和变化环境下的数据库参数优化方法2类.对于固定环境下的参数优化方法,按照方法是否具有从历史任务中学习的能力将研究工作分为传统的参数优化方法和基于机器学习的参数优化方法2类并分别进行介绍.对于变化环境下的参数优化方法,按照不同的变化场景对现有工作进行分类介绍.最后,总结了现有工作中各类方法的优缺点,并对目前研究中待解决的问题和可能发展的方向进行了讨论.Database systems contain a vast number of configuration parameters controlling nearly all aspects of runtime operation.Different parameter settings may lead to different performance values.Parameter tuning can improve the adaptability of database to current environment by selecting appropriate parameter settings.However,parameter tuning faces several challenges.The first challenge is the complexity of parameter space,while the second is the insufficient samples caused by the expensive performance measurements.Moreover,the optimal parameter configuration is not universal when the environment changes.Therefore,regular users and even expert administrators grapple with understanding and tuning configuration parameters to achieve good performance.We summarize and analyze the existing work on parameter tuning for database systems and classify them into two categories:tuning approaches under fixed environments and tuning approaches under changed enviroments,according to whether the approaches have the ability to cope with environmental changes.For the first one,the research work is divided into traditional parameter tuning and machine learning-based parameter tuning according to whether the approaches can learn from historical tasks.For the second one,the existing approaches are introduced according to different environmental change scenarios,respectively.Finally,we summarize the pros and cons of various approaches and discuss some open research problems for parameter tuning.

关 键 词：数据库系统 参数调优 性能调优 机器学习 自治数据库 

分 类 号：TP392[自动化与计算机技术—计算机应用技术]