Database-Operating System Co-Design

We want to investigate how to improve the information flow between a database and an operating system, aiming for better scheduling and smarter resource management. We are interested in identifying the potential optimizations that can be achieved with a better interaction between a database engine and the underlying operating system, especially by allowing the application to get more control over scheduling and memory management decisions. Therefore, we explored some of the issues that arise in database-operating system co-design. Particularly, during the course of the thesis, we ported an existing, main memory, column store engine on top of the Barrelfish OS. Both are research systems developed in the Systems Group at ETHZ addressing the main issues of scalability, hardware heterogeneity and application constraints and requirements. Our extensive analysis shows that the performance of the resulting system and its behavior are stable and resemble the baseline of the columnstore engine's run on top of Linux. Furthermore, the experiments also show that the scale-up version implementation scales up almost linearly with the number of cores. We conclude this report with a discussion that even though currently the column store engine is too simple to be able to fully utilize the scheduling capabilities and resource management features provided by Barrelfish, the two systems form a solid foundation that can be enhanced in the future with a set of more complex features. This will eventually result in a fully functional database engine tightly collaborating with the OS via well defined interfaces