CAJ | 학술논문

基于多核CPU和GPU异构平台的列数据库可用于海量数据和复杂查询，但其优化主要集中在底层，并且后端的执行序列只能通过手工硬编码生成，不能适应多样的SQL查询语句。针对该问题，设计并实现一个将SQL查询语句自动转化成执行序列的编译器，研究多个复杂表达式中的公共子表达式消除和原语依赖图合并方法。与手工编码的比较结果表明，该编译器能够提高算术表达式的计算速度，缩短执行SQL查询语句的时间。
기우다핵CPU화GPU이구평태적렬수거고가용우해량수거화복잡사순，단기우화주요집중재저층，병차후단적집행서렬지능통과수공경편마생성，불능괄응다양적SQL사순어구。침대해문제，설계병실현일개장SQL사순어구자동전화성집행서렬적편역기，연구다개복잡표체식중적공공자표체식소제화원어의뢰도합병방법。여수공편마적비교결과표명，해편역기능구제고산술표체식적계산속도，축단집행SQL사순어구적시간。
A column-oriented database based on a heterogeneous platform of multi-core CPU and GPU can be used for mass data and complex queries. However, the optimization in this database is mainly on physical level and execution sequence for its back-end can only be generated manually, resulting in hard adaptation to varieties of SQL query statements. To solve this problem, this paper designs and implements a compiler that translates SQL query statements into execution sequence. It studies Common Subexpression Elimination(CSE) method in multiple complex expressions and merging method of multiple primitive dependency graphs. Comparing the results with situations where no compiler is used in GSQL, it can be found that this compiler can improve speed of computing multiple complex expressions efficiently and reduce time of processing multiple SQL query statements.