Staged Database Systems

Database system architectures face a rapidly evolving operating environment where millions of users store and access terabytes of data. To cope with increasing demands for performance highend DBMS employ parallel processing techniques coupled with a plethora of sophisticated features. However, the widely adopted work-centric thread-parallel execution model entails several shortcomings that limit server performance, the most important being failure to exploit instruction and data commonality across concurrent requests. Moreover, the monolithic approach in DBMS software has lead to complex designs which are difficult to extend. This thesis introduces a staged design for high-performance, evolvable DBMS that are easy to fine-tune and maintain. I propose to break the database system into modules and encapsulate them into self-contained stages connected to each other through queues. The staged, data-centric design remedies the weaknesses of modern DBMS by providing solutions at (a) the hardware level: it optimally exploits the underlying memory hierarchy, and (b) at a software engineering level: it is more scalable, easier to extend, and more readily fine-tuned than traditional database systems.