Query Processing and Optimization in Modern Database Systems

Abstract

Relational database management systems, which were designed decades ago, are still the dominant data processing platform. Since then, large DRAM capacities and servers with many cores have fundamentally changed the hardware landscape. As a consequence, traditional database systems cannot exploit modern hardware e ectively anymore. This paper summarizes author’s thesis, which focuses on the challenges posed by modern hardware for transaction processing, query processing, and query optimization. In particular, we present a concurrent transaction processing system based on hardware transactional memory and show how to synchronize data structures e ciently. We further design a parallel query engine for many-core CPUs that supports the important relational operators including join, aggregation, window functions, etc. Finally, we dissect the query optimization process in the main memory setting and show the contribution of each query optimizer component to the overall query performance.