内存列存储数据库中优化的混合自适应索引

Optimized Adaptive Hybrid Indexing for In-memory Column Stores

分析型数据库在现代企业中得到广泛应用,在使用过程中对查询处理速度的要求逐渐提高。大数据环境下,分析型数据库面临一系列新的挑战:首先,数据复杂性与日俱增,使得数据库系统的初始配置任务更加繁重,例如索引创建等;其次,在分析过程中,由于查询负载模式无法预知,需要对某些属性反复构建索引,以满足查询的时间要求。显然,传统的索引构建维护技术不能完全满足新的应用环境。数据库分裂技术提出了一种不同的策略去解决这些问题。使用数据库分裂技术,DBA不需要对数据库进行细粒度的系统配置。在查询执行过程中,数据库能自动调整以适应查询负载;随着查询负载的变化,系统自动调整索引。近年来,一系列数据库分裂算法被提出,但已有的算法都各有优缺点。因此给出了一个cache conscious的数据库分裂代价模型,并基于该模型构建了一个新的自适应索引,其可以综合不同数据库分裂算法的优势。通过大量实验验证了这种新自适应索引技术的有效性。

Analytical database has been widely deployed in modern corporations which are posing increasing demand for the speed of data analysis. In the era of big data,analytical database is faced with a number of new challenges. Firstly, the complexity of data keeps increasing, therefore,more efforts have to be put into system configuration, such as index creation. Secondly, without prior knowledge about the patterns of workload, system administrators have to build and re- build indexes repeatedly, in order to meet the time constraints. Apparently, traditional approaches to index construction and maintenance can not work well in the new environment. Database cracking provides an alternative to solve the prob- lem. Using database cracking, a DBA does not need to fine-tune the system configuration. Instead, the database can auto- matically adjust itself to fit the workload during query execution. In recent years, a series of database cracking algo- rithms have been proposed,while none of them is optimal in all situations. The paper proposed a cache conscious cost model for database cracking. Based on the model,we created a new adaptive index, which can combine the advantages of several previous cracking approaches. Extensive experiments were conducted to demonstrate the effectiveness of our method.