学习型数据库索引推荐技术综述

A survey of learning based database index advisor technology

数据库技术是现代计算机软件技术中不可或缺的一部分,在各个行业中都发挥着十分重要的作用。但是,伴随着时代的发展,互联网和云计算要求数据库能够有效地存储和处理大数据,使传统的关系型数据库软件工具能力难以应对新的应用场景。其中,索引对于提升复杂数据集上检索任务的效率有着非常重要的意义。机器学习近年来取得了长足的发展,为数据库索引推荐技术提供了新的机遇。首先叙述了索引推荐的研究背景与挑战;然后,将索引推荐分为索引生成与索引选择2个阶段,分别对这2个阶段的相关技术进行论述;最后,对基于学习的数据库索引推荐技术做了介绍与展望,其中,重点介绍了索引生成与索引选择。索引生成阶段包含基于学习的范围索引、哈希索引、布隆过滤器等,总结了基于学习的索引生成技术的优缺点;对于索引选择阶段,将传统方法与基于学习的方法进行对比与分析,重点叙述强化学习解决索引选择问题。

Database technology is an indispensable part of modern computer software system and plays a very important role in various industries.However,with the development of the times,the Internet and cloud computing require databases to effectively store and process big data,which makes it difficult for traditional relational database software tools to cope with new application scenarios.Among them,indexing is of great significance to improve the efficiency of retrieval tasks on complex data sets.Machine learning has made great progress in recent years,which provides a new opportunity for database index advisor technology.Firstly,this survey summarizes the research background and challenges of index advisor.Then,index advisor is divided into two stages:index generation and index selection.Finally,it summarizes and prospects the database index advisor technology based on learning.Among them,the key techniques are index generation and index selection.The index generation stage includes learning-based range index,hash index and bloom filter,and summarizes the advantages and disadvantages of learning-based index generation technology.For the index selection stage,this survey compares and summarizes the traditional methods with the learning-based methods,and focuses on reinforcement learning to solve the index selection problem.