数据库参数配置智能调优研究综述  

Research Survey on Intelligent Tuning of Database Knobs Configuration

作　　者：李奕言 田季坤 蒲照 李翠平[1,3] 陈红[1,2] LI Yi-Yan;TIAN Ji-Kun;PU Zhao;LI Cui-Ping;CHEN Hong(School of Information,Renmin University of China,Beijing 100872;Engineering Research Center of Database and Business Intelligence,MOE,Beijing 100872;Key Laboratory of Data Engineering and Knowledge Engineering,MOE,Beijing 100872)

机构地区：[1]中国人民大学信息学院,北京100872 [2]数据库与商务智能教育部工程研究中心,北京100872 [3]数据工程与知识工程教育部重点实验室,北京100872

出　　处：《计算机学报》2024年第8期1901-1921,共21页Chinese Journal of Computers

基　　金：国家重点研发计划(2023YFB4503600);国家自然科学基金(U23A20299,62072460,62172424,62276270,62322214)资助.

摘　　要：数据库系统具有大量的参数,这些参数控制了系统的内存分配、I/O优化、备份与恢复等诸多方面,极大地影响着数据库的性能.随着数据库和应用程序的规模和复杂性的增长,传统依靠数据库管理员手动配置参数的方式已经越来越难以满足用户需求.数据库参数配置智能调优将机器学习技术应用到参数调优领域,依据负载信息、数据库参数和性能,借助机器学习算法推荐一组最优的参数.本文针对现有参数配置智能调优技术,从调优方法、应用情况和未来挑战三个方面依次进行梳理和总结.首先将现有参数调优方法依据所用算法不同分为五类,从原理、技术、优缺点等方面对各类方法进行详细介绍和总结.之后介绍当前工业界主流的参数调优工具,分析参数配置智能调优在实际应用过程中遇到的问题及原因.最后,本文对数据库参数配置智能调优的未来研究方向进行了展望.本文旨在帮助研究者掌握当前数据库参数配置智能调优领域主流方法及面临的问题,以推动后续研究工作的开展.The database system encompasses numerous configuration knobs that govern various aspects of its operations.These knobs cover a wide spectrum of database functionalities including memory allocation,I/O optimization,backup and recovery processes.The performance of a database system is heavily influenced by how these knobs are tuned,making their optimization a matter of significant importance.As databases and their associated applications continue to grow in scale and complexity,the conventional approach of manually adjusting these configuration knobs by database administrators is proving to be increasingly inadequate.In response to this emerging challenge,the realm of intelligent database knob tuning has surfaced as a promising solution.This innovative approach leverages machine learning techniques to automate the optimization of database knobs.By analyzing workload information,the settings of the database knobs,and performance metrics,intelligent tuning technologies are capable of recommending an optimal set of knobs that enhance database performance.This paper endeavors to conduct a comprehensive review and synthesis of the current methodologies employed in the intelligent tuning of database knobs.The discourse is structured around three primary focal points:the categorization of existing intelligent knob tuning methods,the examination of prevalent tools within the industry,and an exploration of the future challenges and research directions in this domain.The review begins with a systematic categorization of the existing methods of intelligent knob tuning into five distinct groups,each defined by the underlying machine learning algorithms they employ,including bayesian optimization,reinforcement learning,deep learning,searching based and rule based methods.For each category,a detailed exposition is provided,encompassing the principles,techniques,advantages,and limitations inherent to each method.This categorization not only elucidates the current landscape of intelligent knob tuning but also facilitates a deeper

关 键 词：机器学习 参数调优 贝叶斯优化 强化学习 智能数据库 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]