On the Impact of Hardware on Relational Query Processing

Satisfying the performance needs of tomorrow typically implies using modern processor capabilities (such as single instruction, multiple data) and co-processors (such as graphics processing units) to accelerate database operations. Algorithms are typically hand-tuned to the underlying (co-)processors. This solution is error-prone, introduces high implementation and maintenance cost and is not portable to other (co-)processors. To this end, we argue for a combination of database research with modern software-engineering approaches, such as feature-oriented software development (FOSD). Thus, the goal of this project is to generate optimized database algorithms tailored to the underlying (co-)processors from a common code base. With this, we maximize performance while minimizing implementation and maintenance effort in databases on new hardware. Project milestones: 

• Creating a feature model: Arising from heterogeneous processor capabilities, promising capabilities have to be identified and structured to develop a comprehensive feature model. This includes fine-grained features that exploit the processor capabilities of each device.
• Annotative vs. compositional FOSD approaches: Both approaches have known benefits and drawbacks. To have a suitable mechanism to construct hardware-tailored database algorithms using FOSD, we have to evaluate which of these two approaches is the best for our scenario.
• Mapping features to code: Arising from the feature model, possible code snippets to implement a feature have to be identified.
• Performance evaluation: To validate our solution and derive rules for processor allocation and algorithm selection, we have to perform an evaluation of our algorithms.