Database Analytics: A Reconfigurable-Computing Approach

Bharat Sukhwani, Hong Shick Min, Mathew Thoennes, Parijat Dube, Bernard Brezzo, Sameh Asaad, Donna Dillenberger

IEEE Micro · 2013 · 26 citations · 6 references

Concepts

TL;DR

Businesses need real‑time analytics on growing enterprise data, yet analytics burdens OLTP systems and can degrade their performance. The authors present a highly pipelined, high‑throughput query‑processing engine on an FPGA to offload expensive database analytics queries. The solution enables a DBMS to harness the FPGA accelerator seamlessly, without modifying applications or data layouts. Integrated into a commercial DBMS on an off‑the‑shelf server with a PCIe‑attached FPGA card, the system delivers an order‑of‑magnitude speedup on various real‑life queries.

Abstract

Businesses are experiencing a growing need for performing real-time analytics on ever-increasing enterprise data. While providing useful business insights and improved market responsiveness, analytics also adds a computational burden on traditional online transaction processing (OLTP) systems and could adversely affect the performance of OLTP workloads. The authors present a highly pipelined, high-throughput query-processing engine on a field-programmable gate array (FPGA) to offload expensive queries for database analytics. The proposed solution provides a mechanism for a database management system (DBMS) to seamlessly harness the FPGA accelerator without requiring any changes in the application or the existing data layout. The authors' system, which uses an off-the-shelf server and a PCIe-attached FPGA card and is integrated into a commercial DBMS platform, achieves an order-of-magnitude speedup on various real-life queries.

References

6