机器学习化数据库系统研究综述

Survey on Machine Learning for Database Systems

数据库系统经过近50年的发展,虽然已经普遍商用,但随着大数据时代的到来,数据库系统在2个方面面临挑战.首先数据量持续增大期望单个查询任务具有更快的处理速度;其次查询负载的快速变化及其多样性使得基于DBA经验的数据库配置和查询优化偏好不能实时地调整为最佳运行时状态.而数据库系统的性能优化进入瓶颈期,优化空间收窄,进一步优化只能依托新的硬件加速器来实现,传统的数据库系统不能够有效利用现代的硬件加速器;数据库系统具有成百个可调参数,面对工作负载频繁变化,大量繁琐的参数配置已经超出DBA的能力,这使得数据库系统面对快速而又多样性的变化缺乏实时响应能力.当下机器学习技术恰好同时符合这2个条件:应用现代加速器以及从众多参数调节经验中学习.机器学习化数据库系统将机器学习技术引入到数据库系统设计中.一方面将顺序扫描转化为计算模型,从而能够利用现代硬件加速平台;另一方面将DBA的经验转化为预测模型,从而使得数据库系统更加智能地动态适应工作负载的快速多样性变化.将对机器学习化数据库系统当前的研究工作进行总结与归纳,主要包括存储管理、查询优化的机器学习化研究以及自动化的数据库管理系统.在对已有技术分析的基础上,指出了机器学习化数据库系统的未来研究方向及可能面临的问题与挑战.

As one of the most popular technologies, database systems have been developed for more than 50 years, and are mature enough to support many real scenarios. Although many researches still focus on the traditional database optimization tasks, the performance improvement is little. Actually, with the advent of big data, we have met the new gap obstructing the further performance improvement of database systems. The database systems face challenges in two aspects. Firstly, the increase of data volume requires the database system to process tasks more quickly. Secondly, the rapid change of query workload and its diversity make database systems impossible to adjust the system knobs to the optimal configuration in real time. Fortunately, machine learning may be the dawn bringing an unprecedented opportunity for the traditional database systems to lead us to the new optimization direction. In this paper, we introduce how to combine machine learning into the further development of database management systems. We focus on the current research work of machine learning for database systems, mainly including the machine learning for storage management and query optimization, as well as automatic database management systems. This area has also opened various challenges and problems to be solved. Thus, based on the analysis of existing technologies, the future challenges, which may be encountered in machine learning for database systems, are pointed out.