面向新硬件的可定制数据库系统

Computer hardware is advancing rapidly. On one hand, several types of new storage devices and processors are emerging. On the other hand, computer clusters and cloud platforms are increasingly adopted by applications to handle the growing data volume. Traditional database systems were designed for a particular set of hardware, typically composed of CPU, DRAM and hard disk. Such a monolithic design is not adaptive to the change of hardware, making it difficult for DBMS to benefit from the advance of hardware technologies. This limitation has become a major concern for today's DBMS manufacturers. Besides, the advance of software technology is also changing the landscape of database products. In recent years, specialized database systems have been increasingly favored by users, due to their enhanced performance and usability in the face of particular requirements. General-purpose database systems have to work on their adaptivity and customizability, to preserve their market share. In this project, we rethink the design and architecture of DBMS, and explore methodologies and technologies that can make DBMS customizable to different hardware and applications. The project attempts to conduct three lines of research: first, we will study what DBMS architecture can offer adequate adaptivity to changing hardware; second, we will investigate how to exploit new hardware for efficient data management; third, we will explore the methods to customize a database system to various application requirements.

计算机硬件正处于高速发展期。一方面，新型存储介质和计算芯片不断涌现；另一方面，云化、集群化的硬件平台变得越来越普及。传统数据库系统是针对由CPU、内存和磁盘组成的单一体系结构所设计的系统软件。由于其设计的单一性，它难以适应硬件变化，因而无法充分发挥新硬件的性能优势。这一局限性已成为了数据库厂商的重点关注问题。此外，软件的发展也在逐步改变数据库产品的市场格局。由于软件需求的多样性，面向狭窄应用领域的专用数据库越发凸显在性能和易用性上的优势。为了在市场上保有一席之地，通用数据库系统必须增强自身的适应能力和可定制能力。为了应对以上挑战，本项目致力于研究能够适应不同硬件平台和不同应用需求的可定制数据库系统，提升通用数据库系统的适用范围。项目的研究工作将从三个方面入手：首先，探索更灵活的的数据库系统架构；其次，研究适应未来新型硬件的数据管理技术；最后，探索面向应用的数据库系统的定制化方法。

传统数据库系统由于性能约束而采用了紧耦合的系统架构。这样的架构导致了极高的系统开发成本，并且限制了系统对新的部署平台和新型硬件的适配能力。本项尝试对数据库系统架构进行解耦，并且研究数据库系统对多种新型硬件的适配方法。经过4年的科研工作，项目团队基本证实了数据库系统是能够解耦的，并且能够通过松耦合架构获得更强的适配能力。同时，团队还探索并提出了数据库系统解耦的方法和原则，并且研制了一系列面向新硬件（主要包括持久性内存和RDMA网络）的系统优化技术。项目到解题为止共发表论文11篇，申报专利1项，支撑了5位博士研究生的学位论文工作。