面向内存数据库的逆缓存技术研究

发布时间：2018-07-30 07:38

【摘要】：计算机系统所支持的内存容量不断增加,而内存价格却不断走低,这让内存数据库系统得到了很好的发展时机,它将所有的数据存储于内存中,以此来消除磁盘读写瓶颈。然而在大数据时代,所有数据存储于内存是不可能的,也不是必需的。在理想情况下,数据库系统既想具备内存高效地访问速度,又想具备磁盘的超大容量和低价开销。这关键是如何能同时有效地利用内存和磁盘两者的优势。在本文中,我们对现有最新的解决方案——“逆缓存”进行研究,实现了基于逆缓存的海量键值内存数据库系统原型,并提出了一种更通用和更有效的基于用户态虚拟内存管理(UVMM)的逆缓存方案,能够在小代价的情况下应用于绝大多数内存数据库系统中。在原型系统上,我们组织了大量的实验来对比分析现有方案和UVMM方案的性能,而且研究每种方案的可用性。基于我们的实验发现,UVMM各个方面都表现得很出色,而且在内存限制更高的情况下,UVMM比现有方案拥有更好的性能。最后,我们还总结了一些基本原则,用于设计优秀的逆缓存方案。
[Abstract]:The memory capacity supported by the computer system is increasing and the memory price is decreasing. This makes the memory database system have a good opportunity to develop. It stores all the data in memory to eliminate the bottleneck of disk reading and writing. In the big data era, however, it is impossible and not necessary for all data to be stored in memory. Ideally, the database system not only wants to access memory efficiently, but also wants to have the large capacity and low cost of disk. The key is how to take advantage of both memory and disk effectively. In this paper, we study the newest solution, "inverse cache", and realize the prototype of massive key memory database system based on inverse cache. A more general and effective inverse cache scheme based on user state virtual memory management (UVMM) is proposed, which can be applied to most memory database systems at low cost. In the prototype system, we have organized a large number of experiments to compare and analyze the performance of existing schemes and UVMM schemes, and to study the availability of each scheme. Based on our experiments it is found that the UVMM performs well in every aspect and the performance of the UVMM is better than the existing scheme under the condition of higher memory limit. Finally, we summarize some basic principles for designing excellent inverse caching schemes.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP311.13

【相似文献】