Database Systems and Information Management

Hawk – A Hardware Adaptive Query Compiler

The Hawk project is part of the DFG priority program Scalable data management for future hardware.

A fixed energy budget primarily limits the performance of modern processors. This performance limit forces processor manufacturers to specialize their processors for specific applications in order to achieve the speed increases that users expect. This specialization leads to a landscape of heterogeneous processors, where each processor has unique characteristics and capabilities. Optimizing database systems for processors with different architectures is costly since the database operators have to be reimplemented and optimized for each new processor by experienced programmers.

The Hawk project aims to develop concepts that automatically enable database systems to adapt to such heterogeneous, previously unknown processors. This adaptive optimization avoids manual tuning per processor. The project's core idea is to introduce variations for database operators (e.g., code optimizations, data structures, and parallelization strategies). These variations allow us to create custom per-processor implementations of database operators. Since this variant generation creates a large search space, we will explore concepts for a query optimizer that can automatically choose efficient implementations of an operator for each processor without prior knowledge of the processors used.

Further information can be found at https://www.dfg-spp2037.de/ma4662-5/.

 

Project Duration:29/05/2017 - 30/04/2022

Funding Agency: Deutsche Forschungsgemeinschaft (DFG)

Project Partner: Deutsches Forschungsentrum für Künstliche Intelligenz (DFKI)