Funded by:

• Zentrales Innovationsprogramm Mittelstand (ZIM) des Bundesministeriums für Wirtschaft und Eneregie (BMWi) – Fördermodul Kooperationsprojekte

Runtime:

• 2013 - 2015

Project Coordinator:

• Prof Dr. rer. nat. habil. Sven Groppe
• Prof. Dr. Thilo Pionteck

Research Associate:

• Dipl. Inf. Stefan Werner

Motivation

Relational databases with large data volumes causes performance issues and have a negative user experience.

Goals and Procedure

The goal of this project is hardware acceleration of database operations in relational databases with large data volumes. This addresses the core issue of so-called “big data” scenarios by enabling database access and thus data processing to be accelerated without requiring major interventions in the customers’ data and analysis tools.

Innovation and Prospects

The technical implementation and innovation should be achieved using a combined hardware/software system that outsources time-consuming database operations to a scalable cluster of programmable logic devices (FPGAs, Field Programmable Gate Arrays). The FPGA-Cluster DB database management system to be developed is intended to process several queries at the same time, recognize time-consuming operations and outsource them to a cluster of FPGAs. The FPGAs only ever implement the circuit structures that are required for the requests currently being processed. This is achieved using partial dynamic reconfiguration, which makes it possible to change parts of the configuration of an FPGA and thus its implemented circuit structures at runtime. The currently required data paths are assembled from a set of pre-synthesized database operators according to the modular principle and written into the FPGA cluster.

In this context, the IFIS is working on the subproject "Integration of hardware accelerators into database systems", which deals with the process and database-related aspects of the project. The widely used open source database system MySQL is to be expanded so that the possibilities of hardware accelerators are taken into account when creating the operator graph and the corresponding operators are outsourced to the FPGA cluster. The optimization potential should also be exploited when processing parallel requests.

Project Partners

• Institute of Information Systems (IFIS), University of Lübeck
• Technical University of Dresden

Dissertation

Stefan Werner:
Hybrid Architecture for Hardware-accelerated Query Processing in Semantic Web Databases based on Runtime Reconfigurable FPGAs
eingereicht Oktober 2016, angenommen Januar 2017, mündl. Prüfung am 06.02.2017 (Doktorvater: Groppe), Thesis