A Novel Database Acceleration Technology for Full Table Scans

Chuang Zhang, Xuebin Tang, Yuanxi Peng

Abstract: Advancements in hardware technology, especially Field Programmable Gate Arrays (FPGAs), have shown potential in enhancing database performance. Yet, much of the existing research has neglected crucial aspects such as the database-hardware interface and operator scheduling management. This oversight limits the capability of database systems to efficiently process complex SQL statements, thereby constraining their functionality. This paper addresses these issues by introducing a comprehensive full-stack design architecture that integrates database systems with FPGA technology. The proposed architecture emphasizes: 1) Supporting native database storage formats and complex filtering conditions through the re-encoding of SQL statements; 2) Implementing an innovative out-of-order scheduling strategy to enable parallel processing and efficient data transmission; 3) Developing an FPGA-based micro-architecture tailored for enhanced data scanning and dynamic filter expression calculations. Extensive experimental evaluations demonstrate significant performance enhancements, including a threefold increase in processing speed and a 34% reduction in CPU load. These improvements underscore the effectiveness of our proposed software-hardware cooperative optimization approach in accelerating database operations.