The role of flash memory in database management systems

or the first time in the history of data processing systems, traditional hard disk drives (HDDs) are under pressure from flash storage in the form of solid state drives (SSDs) as the home for persistent data. SSDs fill the big gap in latency between the main memory and the HDDs, especially for random I/O accesses. However, the internals of database management systems (DBMSs) are traditionally designed and optimized for HDDs -- for example, DBMS query processing methods go to great lengths to employ mechanism that generate sequential I/O accesses to the underlying I/O storage subsystem. The focus of this dissertation is on exploring how SSDs can be effectively integrated into existing DBMSs to improve the performance of database query processing methods.;The first part of this thesis examines how traditional join algorithms are impacted when the database is stored entirely in SSDs. The next component of this thesis deals with extending the DBMS buffer pool to gracefully spill evicted pages from the main-memory database buffer pool into a SSD-based cache. In that architecture HDDs are still used to store the database, and the DBMS employs the SSD to accelerate the DBMS query processing performance. This approach allows building database configurations in which one can balance the amount of SSD, which is typically more expensive than HDDs from the...