Scalable Analytical Query Processing

TL;DR: This work proposes to avoid possibly wrong optimizer decisions regarding physical join operators by replacing these operators by a single one, called g-join, and develops a suite of massively parallel sort-merge (MPSM) join algorithms, which exploit the parallelization potential of multi-core CPUs.

Abstract: Analytical query processing in database management systems aims at providing information within an acceptable time while affecting the performance of concurrent transactional workloads as little as possible. Scalability refers to the ability of database management systems to take advantage of additional resources to improve performance. In order to achieve the goal of scalable analytical query processing, we examine three approaches: First, we focus on synergy-based workload management, which exploits synergies between concurrently executed queries in order to maximize performance. Second, we concentrate on the robust execution of single queries. We propose to avoid possibly wrong optimizer decisions regarding physical join operators by replacing these operators by a single one, called g-join. Third, we turn our attention toward modern architectures and develop a suite of massively parallel sort-merge (MPSM) join algorithms, which exploit the parallelization potential of multi-core CPUs.