内存数据库存储管理研究
发布时间：2018-11-26 08:50
【摘要】：内存数据库由于全部或者大部分数据常驻内存，故其事务处理过程中的I/O很少。因此，与磁盘数据库以减少I/O为主要优化目标不同，内存数据库不再以此为主要优化目标。这也意味着磁盘数据库中的各种处理方法不再适用于内存数据库，而需要针对内存数据库的自身特点来研究新的处理方法。作为内存数据库的各种功能的基础，存储管理有着十分重要的地位。存储管理不仅直接影响着内存数据库的性能，也影响了各种上层功能，故内存数据库存储管理的研究具有重要的意义。 基于以上需求，研究了内存数据库的存储管理，，并给出了相应的存储管理器的设计与实现。具体而言，研究内容包括内存数据库的数据组织、内存管理以及索引。主要工作如下：在分析内存数据库的存储层次的基础上，设计了存储管理器的存储层次以及记录的组织方法；通过改进记录的格式从而大幅度提高了存储管理器的性能；设计并实现了一个基于多内存池，可以有效利用内存资源的内存管理方法；设计并实现了一个灵活的启动加载策略和一个内外存数据交换策略；详细分析了哈希索引、T-树索引和缓存敏感索引的特点及其适用性，给出了T-树索引的详细设计、实现及其优化方法。 最后，通过实验表明，索引节点块的大小对索引性能存在着较大的影响；记录格式的改进显著提升了该存储管理器的性能。另外，还通过与FastDB的对比实验表明，该存储管理器的综合性能较实际内存数据库系统FastDB相当或更高，在实践中是可行的。
[Abstract]:Since all or most of the data are resident in memory, there are very few I / O in the transaction. Therefore, unlike the disk database with reducing I / O as the main optimization goal, the memory database is no longer the main optimization goal. This means that the various processing methods in the disk database are no longer suitable for the memory database, but the new processing method should be studied according to the characteristics of the memory database. As the basis of various functions of memory database, storage management plays an important role. Storage management not only directly affects the performance of the memory database, but also affects various upper functions, so the research of memory database storage management is of great significance. Based on the above requirements, the storage management of the memory database is studied, and the design and implementation of the storage manager are given. Specifically, the research includes data organization, memory management and index of memory database. The main work is as follows: on the basis of analyzing the storage hierarchy of the memory database, the storage layer of the storage manager and the organization method of the record are designed, and the performance of the storage manager is greatly improved by improving the format of the record. A memory management method based on multi-memory pool is designed and implemented, and a flexible startup loading strategy and an internal and external memory data exchange strategy are designed and implemented. The characteristics and applicability of hash index, T- tree index and cache sensitive index are analyzed in detail. The detailed design, implementation and optimization of T- tree index are given. Finally, the experimental results show that the size of the index node block has a great impact on the index performance, and the improvement of the record format significantly improves the performance of the storage manager. In addition, the comparison experiment with FastDB shows that the comprehensive performance of the storage manager is comparable to or higher than that of the real memory database system FastDB, and it is feasible in practice.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP311.13;TP333
本文编号：2358031
[Abstract]:Since all or most of the data are resident in memory, there are very few I / O in the transaction. Therefore, unlike the disk database with reducing I / O as the main optimization goal, the memory database is no longer the main optimization goal. This means that the various processing methods in the disk database are no longer suitable for the memory database, but the new processing method should be studied according to the characteristics of the memory database. As the basis of various functions of memory database, storage management plays an important role. Storage management not only directly affects the performance of the memory database, but also affects various upper functions, so the research of memory database storage management is of great significance. Based on the above requirements, the storage management of the memory database is studied, and the design and implementation of the storage manager are given. Specifically, the research includes data organization, memory management and index of memory database. The main work is as follows: on the basis of analyzing the storage hierarchy of the memory database, the storage layer of the storage manager and the organization method of the record are designed, and the performance of the storage manager is greatly improved by improving the format of the record. A memory management method based on multi-memory pool is designed and implemented, and a flexible startup loading strategy and an internal and external memory data exchange strategy are designed and implemented. The characteristics and applicability of hash index, T- tree index and cache sensitive index are analyzed in detail. The detailed design, implementation and optimization of T- tree index are given. Finally, the experimental results show that the size of the index node block has a great impact on the index performance, and the improvement of the record format significantly improves the performance of the storage manager. In addition, the comparison experiment with FastDB shows that the comprehensive performance of the storage manager is comparable to or higher than that of the real memory database system FastDB, and it is feasible in practice.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP311.13;TP333
【参考文献】
相关期刊论文 前5条
1 刘云生,李国徽,肖群;实时内存数据库的数据交换策略及评价[J];计算机研究与发展;1999年07期
2 刘云生,迟岩;内存受限的实时内存数据库数据装入策略[J];计算机工程;2004年20期
3 郭超;李坤;王永炎;刘胜航;王宏安;;多核处理器环境下内存数据库索引性能分析[J];计算机学报;2010年08期
4 王珊;肖艳芹;刘大为;覃雄派;;内存数据库关键技术研究[J];计算机应用;2007年10期
5 刘云生,李国徽;实时内存数据库的装入[J];软件学报;2000年06期
本文编号：2358031
本文链接：https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/2358031.html
