Research And Implementation Of SQL Compiler On Column-oriented Database

With the information society stepping into the age of big data, the databases are required to have the higher performance for queries because of large amounts of data. Owning to the influence of IO, the promotion space of traditional row-based database islimited. In order to deal with mass data query bottleneck, the team research and design a column-based data query system. Unlike row-based relational database, the system inroduces the concept of column store into data storage, and uses parallel computing technology, applying various optimizational methods to accelerate the primitives and query plan on the MIC and multi-core CPU heterogeneous platform. This system can only support part of style in SQL statement queries before, unable meeting the diversity of complex SQL query.Therefore, this paper implements a more complete database compiler based on previous work, which can support more style of complex queries. The parsing process is divided into three steps: the first step is converting the SQL query to a syntax tree through syntax and morphology analysis; the second step is to generate a logic query plan from a syntax tree using a plan generator, which is constituted by 8 table operators; the third step is to form the final primitive dependence graph by resolving the logic operator node. Considering of the characteristics of column-based database, this paper design multi-condition filtering primitive and index-based loading primitive, which may shorten the length of primitive execution list and save the cost of disk I/O. This paper will also give the solutions about the optimizational scheme of relevant and non-relevant sub-queries.Through the test of 22 complex queries provided by international testing agency, the database compiler in this paper can support the correct analysis of 15 complex queries, and after comparing the time consuming, it has better efficiency after the optimization of query.