MC-DQA: An Efficient Database Query Acceleration Architecture Based on FPGA

The data size and concurrent query probability involved in databases are growing rapidly, which leads to additional SQL query delay overhead, and it is becoming challenging to realize efficient database SQL query through FPGA heterogeneous design. In order to reduce the impact of this bottleneck on database performance, a novel Multi-card Collaborative Database Query Architecture (MC-DQA) based on FPGA is proposed to meet the requirements of high-performance database query processing. Unlike traditional heterogeneous acceleration methods, our architecture adopts a hardware design form of “single database server carrying multiple heterogeneous acceleration cards”. CPU executes MC-DQA scheduling strategy to allocate SQL query tasks to various acceleration cards, the cards and kernels parallel acceleration calculation cooperate, thus reducing the query processing time for high concurrency large-scale data. The MC-DQA we designed can be applied to current mainstream relational database management systems. On KaiwuDB, the SQL query experiment based on TPC-H benchmark and Inspur dataset proved the advantages and feasibility of our design architecture. Compared with the traditional “CPU-FPGA” scheme, MC-DQA can save up to 72.29% of query time overhead.