Increasing resource demands require relational databases to scale. While relational databases are well suited for vertical scaling, specialized hardware can be expensive. Conversely, emerging NewSQL and NoSQL data stores are designed to scale horizontally. NewSQL databases provide ACID transaction support; however, joins are limited to the partition keys, resulting in restricted query expressiveness. On the other hand, NoSQL databases are designed to scale out linearly on commodity hardware; however, they are limited by slow join performance. Hence, we consider if the NoSQL join performance can be improved while ensuring ACID semantics and without drastically sacrificing write performance, disk utilization and query expressiveness. This paper presents the Synergy system that leverages schema and workload driven mechanism to identify materialized views and a specialized concurrency control system on top of a NoSQL database to enable scalable data management with familiar relational conventions. Synergy trades slight write performance degradation and increased disk utilization for faster join performance (compared to standard NoSQL databases) and improved query expressiveness (compared to NewSQL databases). Experimental results using the TPC-W benchmark show that, for a database populated with 1M customers, the Synergy system exhibits a maximum performance improvement of 80.5% as compared to other evaluated systems.
翻译:虽然关系数据库非常适合垂直缩放,但专门硬件可能非常昂贵。相反,新兴的NewSQL和NOSQL数据存储是横向设计的。新SQL数据库提供ACID交易支持;然而,合并仅限于分区键,导致有限的查询表达性。另一方面,NoSQL数据库的设计是为了缩小商品硬件的线性管理;然而,它们由于加入性能缓慢而受到限制。因此,我们认为,新SQL数据库的合并性能能否在确保ACID语义和不大幅牺牲写作性能、磁盘使用和查询直观性能的同时得到改进。本文介绍了协同系统,该系统利用Schema和工作量驱动的机制,在NOSQL数据库的顶端确定实际化观点和专门的共通制控制系统,以便能够根据熟悉的关系公约进行可扩展的数据管理。协同交易略微写性性能退化和增加磁盘利用率,以便更快地合并性能(与标准 NoSQL数据库相比),以及改进的查询显性性(与NewSQL数据库相比,磁盘利用率和工作量驱动力数据库相比),实验结果显示80个用户的SyW数据库。