Scalable Filtering Modules for Database Acceleration on FPGAs
Kristiyan Manev, Anuj Vaishnav, Charalampos Kritikakis, Dirk Koch · 2019
Database sizes are growing faster than the processing power in the post-Moore era due to the advent of big data applications, which make hardware acceleration mandatory. However, currently, database acceleration using FPGAs has mainly been static and with limited accelerator functionality, reducing the potential performance gains from customization using FPGAs. In this paper, we propose a dynamic stream processing architecture for SQL query execution on FPGAs. It achieves this by building pipelines based on scalable database accelerator primitives and partial reconfiguration. Further, we introduce novel optimization techniques to design a scalable filtering module for database restriction and Boolean evaluation. It features multiple PEs that operate in parallel and implements DNF solver to accelerate Boolean expression evaluation. Our evaluation shows that not only the system can support the acceleration of filtering in all TPC-H queries but provides up to 17.7GB/s throughput and scales linearly with the data-path size.