FPGA Based Database Sort-Aggregation Query Acceleration Architecture

FPGA Based Database Sort-Aggregation Query Acceleration Architecture

연구 분야: Databases

학회: 2024 9th International Symposium on Computer and Information Processing Technology (ISCIPT)

In AI era, the scale of data involved in relational databases has shown an exponential growth trend, which leads to a sharp rise in the CPU load of database management systems and a significant increase in SQL query latency. Improving the response efficiency of databases to SQL queries to massive storage resources through heterogeneous hardware optimization design has become a research hotspot in related fields. This work proposes an FPGA based database sort-aggregation acceleration architecture that meets the requirements of contemporary high-performance database applications. The design introduces heterogeneous FPGAs to offload high concurrency large-scale SQL sort-aggregation query tasks that would need to be executed by CPU. The acceleration kernels included in the architecture is based on practical relational database SQL queries application, and an iterative merge-sort kernel and corresponding aggregation query kernel design are implemented through FPGA logic. In addition, the architecture also supports multi-threaded concurrent processing to maximize the FPGA's excellent parallel capabilities and improve the efficiency of SQL queries. Experiments on the open-source database KaiwuDB show that the FPGA based sort-aggregation acceleration architecture can reduce the average time cost by 88.26% compared with the database CPU independent query scheme.