Accelerating Database Analytics Through FPGAs

Speaker: Osman Unsal (BSC)
Date: Thursday, 3 December 2015, 15:30-17:30
Session: FPGAs
Talk type: Short talk (15 min)

Abstract: In this talk, we will explore how to accelerate Database Analytics through the use of FPGAs. As a side issue, we will also discuss how to bring these specialized accelerators closer to storage devices to do preprocessing of data. We have run full and complex SQL queries directly in the FPGA fabric, with the fabric being directly connected to SSD disks through the SATA interface. We believe that the experiennce we gained through this exercise could be applicable to HPC systems as well. Our proposed FPGA platform includes novel and specialized database acclerators for frequent Database operators such as filtering, aggregation, sorting and hash-joins. In order to test the design, we executed TPC-H Query in 1GB and 10GB configurations. We compared our results with PostgreSQL 9.2 running on a 32-core Intel Xeon E5-2670 server. In 1GB scale, execution time is reduced by up to 17.7x in 1GB scale and by up to 90x in 10GB scale, compared to the baseline.