Many-core architectures for database acceleration

Authors

Kai-Uwe Sattler, Jens Teubner, Felix Beier, and Sebastian Breß

Published

Proc. Database Systems for Business, Technology and Web (BTW), Hamburg, Germany, March 2015.

Abstract

Physical and technological limitations in increasing the clock frequency of processors have accelerated the development of multi- and many-core architectures in recent years. However, the utilization of these architectures requires an extensive parallelization of computations. For the database sector, this means that established data structures and procedures need to be rethought, but at the same time opens up new possibilities for accelerating and scaling database processing. The aim of the tutorial is therefore to provide an overview of the current state of research and the possible applications of many-core architectures in database systems. In addition to standard processors, the focus will be on GPGPU architectures, which already enable the use of thousands of cores. Based on a presentation of current many-core and GPU architectures and the programming models and APIs based on them, a series of representative fields of application will be discussed. In addition to standard database operators, indexing of data and the implementation of user-defined functions (UDFs), possibilities for supporting DBMS components such as the optimizer are also covered. A second focus is on aspects of heterogeneous platforms consisting of CPU and GPU, such as placement and data transfer issues. The target group of the tutorial are database researchers and developers who are interested in the use of modern hardware platforms in database systems and would like to find out about the current state of research. You will gain an insight into programming models and techniques as well as decision support for the use of many-core and GPU architectures for database acceleration.